20201009_RACER_Proposers_Day_.pdf

UNCLASSIFIED

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER

Robotic Autonomy in Complex Environments with Resiliency

Proposers Day

October 9, 2020

Distribution Statement A – Approved for Public Release, Distribution Unlimited 2

Agenda – RACER Virtual Proposers Day

0900 Welcome Dr. Stuart Young RACER PM, DARPA TTO 0900-0910 DARPA RACER Security Overview Mr. Jason Webber DARPA SID/TTO PSR 0910-0930 RACER BAA – CMO Overview Mr. Ovidio Gonzalez-Nunez DARPA CMO 0930-0945 Introduction & Comments – DARPA TTO Dr. Mike Leahy Director, DARPA TTO 0945-1030 RACER Program Introduction Dr. Stuart Young RACER PM, DARPA TTO 1030-1100 Break 1100-1130 NGCV & Robotic Autonomy COL Warren Sponsler Chief of Staff, NGCV CFT 1130-1200 PM-RCV & NGCV CFT RCV Program LTC Chris Orlowski PEO-GCS PM-RCV 1200-1245 Overview/Intent of RACER Phase 1 & BAA Dr. Stuart Young RACER PM, DARPA TTO 1245-1330 ARL 6.1/6.2 Autonomy Stacks (RCTA, etc.) Dr. Ethan Stump ARL AI Maneuver/Mobility

(RACER Stack Baseline, Code Repository) Program Chief Scientist Dr. Jon Fink ARL AI Maneuver/Mobility

Mobility Lead 1330-1400 Complementary GVSC Activities Dr. David Gorsich GVSC Chief Scientist 1400-1430 Q&A Dr. Stuart Young RACER PM, DARPA TTO

Distribution Statement A – Approved for Public Release, Distribution Unlimited 3

Security Overview Jason Webber, DARPA SID/TTO PSR

Distribution Statement A – Approved for Public Release, Distribution Unlimited 4

RACER BAA – DARPA Contracts Management Office Mr. Ovidio Gonzalez-Nunez, DARPA CMO

• BAA Issuance (https://beta.sam.gov): October 13, 2020

• Abstract Due Date: October 28, 2020, 4:00pm Eastern Time

• FAQ/Questions Due Date: November 13, 2020, 4:00pm Eastern Time

• Full Proposal Due Date: December 11, 2020, 4:00pm Eastern Time

• BAA Closing Date: 180 days from date of posting on beta.SAM.gov

Key Milestones

NOTE: If there is any discrepancy between what is presented today and the BAA, the BAA takes precedence.

5UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Described in Federal Acquisition Regulation Part 35:

For the acquisition of basic and applied research FAR 35.016 (a): Shall only be used when “meaningful proposals with varying technical/scientific approaches can be reasonably anticipated.”

Not a Request for Proposal (RFP) FAR Part 15 does not apply*

BAA Process: Regulations

*NOTE: FAR SUBPART 15.4 WILL APPLY FOR CONTRACT PRICING

6UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Multiple Awards are anticipated

• At this time, DARPA is soliciting full proposals for Phase 1 only:

Phase 1 shall comprise of:

An 18-month Phase 1 base period A 3-month Phase 1 option

Proposers shall also provide an initial program plan, initial schedule, and Rough Order of Magnitude (ROM) cost for a 30-month Phase 2 option

DARPA plans to select 1 or more performers to continue on to Phase 2 Selection of Phase 2 performers will be limited to Phase 1 prime contractors Updated proposal guidance and a request for Phase 2 proposals will be provided to Phase 1 performers toward the end of Phase 1

Program Structure

7UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Clearly mark proprietary and classified information

• Volume I, Technical and Management Proposal:

Section I: Administrative Section II: Summary of Proposal Section III: Detailed Proposal Information

• Volume II, Cost Proposal (must be unclassified):

Cost tables in MS Excel format w/ formulas intact Cost breakdown: Direct labor, indirect rates, ODC, Material Subcontractor proposals: Required; may be submitted to Govt. directly (via same submission methods as those of prime) Supporting documentation for proposed ODC & Material amounts

Proposal Requirements

8UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Prospective proposers are strongly encouraged to submit abstracts

• Format is a greatly condensed version of full proposal format

• 5-page maximum

• DARPA will respond with a statement as to encourage or discourage a proposal submission

• Proposal may be submitted irrespective of DARPA’s abstract response

• Proposal evaluation will be based solely on proposal’s merit against evaluation criteria and without regard to abstract feedback

Abstracts

9UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Per FAR 35.016(d): “Proposals received as a result of the BAA shall be evaluated in accordance with evaluation criteria specified therein through a peer or scientific review process. Written evaluation reports on individual proposals will be necessary but proposals need not be evaluated against each other since they are not submitted in accordance with a common work statement.”

• Proposals are not ranked. No color, adjectival, or numerical “scoring” systems are employed during the Scientific Review Process

Evaluation

10UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• In Descending Order of Importance:

Overall Scientific and Technical Merit Proposer’s Capabilities and Related Experience Potential Contribution and Relevance to the DARPA Mission Realism of Proposed Cost and Schedule

• Complies with guidance at FAR 35.016(e):

“The primary basis for selecting proposals for acceptance shall be technical, importance to agency programs, and fund availability. Cost realism and reasonableness shall also be considered to the extent appropriate.”

Evaluation Criteria

11UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• DARPA reserves the right(s) to:

Select for negotiation all, some, one, or none of the proposals received and to make awards with/without discussions

Accept proposals in their entirety or only portions of proposals for award Segregate portions of proposal into pre-priced options Negotiations may be opened with proposer

Award Information

12UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• DARPA reserves the right(s) to:

Fund proposals in phases with options for continued work at the end of one or more of the phases

Remove proposers from award consideration should the parties fail to reach agreement on award terms, conditions and cost/price

• Fundamental/Non-Fundamental Research:

Proposers must indicate whether they believe the scope of their proposed research constitutes Fundamental Research

Award Information, cont’d.

13UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• FAR-based procurement contract, Grant, Cooperative, Agreement, or Other Transaction (OT) Agreement

• Proposers without a DCAA-approved cost accounting system should submit an SF 1408 with proposal to receive a cost reimbursement contract

• Other Transaction (OT) – must meet eligibility criteria – refer to 10 U.S.C. § 2371b

• CO has sole discretion to select award instrument type and to negotiate all instrument terms and conditions with selectees

Award Instrument(s)

14UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• All responsible enterprises capable of satisfying the Government’s needs

• Non-U.S. organizations/individuals may participate to the extent that such participants comply with any necessary non-disclosure agreements, security regulations, export control laws and other governing statutes applicable under the circumstances.

• No portion of the BAA will be reserved for small business, no evaluation preference

• Classified submissions have specific requirements

Eligibility Information

15UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Government agencies/labs, FFRDCs cannot propose to this BAA in any capacity, UNLESS:

Can clearly demonstrate the work is not otherwise available from the private sector, AND

Provide written documentation citing the specific statutory authority (as well as, where relevant, contractual authority) establishing eligibility to propose to government solicitations

Eligibility Information, cont’d.

16UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Organizational Conflicts of Interest:

Without prior approval or a waiver from the DARPA Deputy Director, in accordance with FAR 9.503, a contractor cannot simultaneously provide scientific, engineering, technical assistance (SETA) or similar support and also be a technical performer

Must address in your proposal if providing SETA or similar support to any DARPA technical office(s) through an active contract or subcontract

Contact DARPA in advance of submitting proposal

Eligibility Information, cont’d.

17UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Small Business Subcontracting Plan (required for proposals for FAR contracts > $700K by other than small businesses)

• Certified Cost and Pricing Data Required (FAR contracts > $2M)

• Online Representations & Certifications – FAR & DFARS NAICS 541715 – R&D in the Physical, Engineering, and Life Sciences

• Wide Area Workflow (WAWF)

Proposal Considerations

18UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Export Control – Clause/Language will be included in award

• Subcontracting

• System for Award Management (SAM)

• Cost Accounting Standards (CAS) Notice & Cert

• Reporting Executive Compensation and First-Tier Subcontract Awards

• Safeguarding Covered Defense Information and Cyber Incident Reporting

Administrative & Policy Requirements

19UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Data Rights Assertions – DFARS 252.227-7013/7014:

Identify all non-commercial and commercial technical data & computer software to be generated, developed, and/or delivered to which the Government will receive less than Unlimited Rights and assert specific restrictions on those deliverables

Assertions required for Prime and Subs Use defined “Basis of Assertion” and “Asserted Rights Category” Justify “Basis of Assertion” This information is assessed during evaluations and will be incorporated into any contract award Avoid broad/vague assertions

Intellectual Property

20UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• PATENTS: Include documentation proving your ownership of or possession of appropriate licensing rights to all patented inventions (or inventions for which a patent application has been filed) that will be utilized under your proposal for the DARPA program.

• i-Edison: All patent reports and notifications must be submitted electronically

Intellectual Property, cont’d.

21UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Please refer to guidance in BAA regarding how to submit questions

• After Receipt of Proposals – Government (PM/PCO) may communicate with proposers to understand the meaning of some aspect of the proposal that is not clear or to obtain confirmation or substantiation of a proposed approach, solution, or cost estimate

• Only a duly authorized Contracting Officer may obligate the Government

• Informal feedback will be provided upon request once selections are made. No formal “debriefing” (see FAR 15.506) will be provided.

Communications

22UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Read the BAA in its entirety – several times

• Do not wait until the last hour to submit the proposal, as the website may experience latency due to site traffic

• Fully support all costs proposed – err on the side of providing too much supporting documentation

• Write the proposal with the evaluation criteria in mind

• Be detailed and thorough

“Best Practices”

23UNCLASSIFIED Distribution Statement A – Approved for Public Release, Distribution Unlimited

Distribution Statement A – Approved for Public Release, Distribution Unlimited 6

Introduction & Comments – DARPA TTO Dr. Michael Leahy, Director, TTO

Distribution Statement A – Approved for Public Release, Distribution Unlimited 25

RACER Program Introduction Dr. Stuart Young, RACER PM, DARPA TTO

Distribution Statement A – Approved for Public Release, Distribution Unlimited

Robotic Autonomy in Complex Environments with Resiliency Dr. Stuart Young

Program Manager, DARPA TTO

Proposers Day Program Overview

October 9, 2020

RACER

Distribution Statement A – Approved for Public Release, Distribution Unlimited 27

Questions:

• Please submit your written questions via the Zoom question feature

• This will allow the DARPA RACER team to capture the Question and Answer session for those unable to attend

• Answers to the questions will be provided later and posted to https://beta.sam.gov/

• Presentation materials may be made available after clearing public release

Proposers Day vs. BAAs / Questions

Note:

In the event of a BAA posting, if there is any discrepancy between what is presented today and the BAA, the BAA takes precedence

RACER - Autonomy to Drive Off-Road at Speed

Objective Area (Red Force)

Direct Transition to Robotic Combat Vehicle Programs

Game Changing System Demonstrations

Off-Road Virtual Development Environments

RACER provides foundational autonomous mobility for Intelligent Combat Maneuver by 2030

UNCLASSIFIED

Distribution Statement A – Approved for Public Release, Distribution Unlimited

Distribution Statement A – Approved for Public Release, Distribution Unlimited 29

RACER End State Off-Road Autonomous Driving and Simulation Redefined

Game changing system demonstrations

• Autonomous mobility in unstructured off-road terrain

• Frequent platform-based, DARPA-hosted field experimentation to prove performance

• Combination of platform-based and simulation-based development

Advance off-road virtual development environments

• Autonomy modeling and simulation tools for multiple environments

• Virtual environments tightly coupled with sensing physics, vehicle dynamics, and terrain

• Simulation-developed autonomy with intent to compare sim-to-real environment performance

Direct transition to Robotic Combat Vehicle Program

• Compatible autonomy algorithms, a Government Purpose Rights code repository, off-road autonomy simulation toolsets, and experienced industry teams

• Removes autonomous mobility as the pacing limit to speed on the battlefield

• Enables new unmanned or existing optionally manned vehicles

Distribution Statement A – Approved for Public Release, Distribution Unlimited 30

U.S. Army Autonomous Platform Demonstrator (APD)

• Ground-up combat vehicle design for unmanned operation

• Evolved from DARPA TTO UGCV and UPI Programs

APD Platform Attributes

• High Speed 105 km/h

• Combat Platform Scale 9 tons

• Extreme off-road mobility 1m gap/step

• Extreme platform resiliency

APD Autonomy

• Blind (no perception)

• Remote Control (RC), Teleoperation, or Waypoint Following only, and all a with human operator

Capable Platforms Exist; Autonomy Does Not

Distribution Statement A – Approved for Public Release, Distribution Unlimited

Platform-Based Development Off-road maneuver performance on autonomous combat-scale platforms

Two Development Approaches Two Autonomy Algorithm/Stack

Development Approaches One Demonstration Objective

Field Develop-test-develop-test

Cycles

Complex terrain, long distances at autonomous maneuver speeds

Open terrain, short distances at autonomous speeds of mechanical limits

Multiple platforms at local test sites

Distribution Statement A – Approved for Public Release, Distribution Unlimited 32

Autonomous Mobility in Unstructured Off-Road Terrain

Short Range Adequate 3D knowledge of the immediate vicinity around the platform

Medium Range 2D understanding of the world within sensor range but at the extents of or just beyond 3D range

Long Range Global routes in the transition between 2D understanding to beyond all sensors

Short

Medium

Long

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER Development Plan

Distribution Statement A – Approved for Public Release, Distribution Unlimited 34

RACER Performance Targets

UPI/Crusher (2008)

Legged Squad Support System

(2013)

Squad Mission Support System

(2010)

PerceptOR (2004)

Pacing Region for Maneuver

Forces

Pacing Region for Reaching APD

Platform Limits

RACER end state in complex terrain

RACER end state in open terrain

Legend (Axis Platforms) Manned Unmanned

APD

Max.

Off-Road SpeedOff-Road State of the Art

(Avg. Autonomous Speeds Achieved)

Distribution Statement A – Approved for Public Release, Distribution Unlimited 35

DARPA-Hosted Field Experiments Demonstration Approach - Supports the Development of Resilient Autonomy

Autonomy capable of adapting to varied terrains, including novel environments

Run

Speed Run 1 (~5km)

UGV

Complex, long distance Run 1

Metrics – Avg. Auto/Interv.

Run 1 (15-30km)

UGV

Complex, long distance Run 2

Metrics – Avg. Auto/Interv Run 2 (15-30km)

UGV w/Tactics Complex, long distance Run 3

Metrics – Avg. Auto/Interv Run 3 (15-30km)

UGV w/Tactics Complex, long distance Run 4

Metrics – Avg. Auto/Interv Run 4 (15-30km)

Speed Run 2 (~2-5km)

Complex, long distance Open, short distance High speed @ platform limits

Distribution Statement A – Approved for Public Release, Distribution Unlimited www.darpa.mil

Distribution Statement A – Approved for Public Release, Distribution Unlimited 37

Break

Distribution Statement A – Approved for Public Release, Distribution Unlimited 38

NGCV & Robotic Autonomy COL Warren Sponsler, Chief of Staff, NGCV CFT

Distribution Statement A – Approved for Public Release, Distribution Unlimited 39

PM-RCV & NGCV CFT RCV Program LTC Chris Orlowski, PhD, PEO-GCS / PM-RCV

Use or disclosure of data contained on the page is subject to restrictions on title page.

DISTRIBUTION STATEMENT A. Approved for public release, distribution unlimited.

UNCLASSIFIED

UNCLASSIFIED

U.S. Army Robotic Combat Vehicles Overview

RACER Proposers’ Day

LTC Christopher Orlowski, PhD | 09 OCT 2020 Product Manager, Robotic Combat Vehicles

Use or disclosure of data contained on the page is subject to restrictions on title page.

What are Robotic Combat Vehicles (RCVs)?

• RCVs are a potential family of vehicles (Light, Medium, Heavy) that enhance the capabilities of U.S. Army Brigade Combat Teams conducting Cross-Domain Maneuver in support of Joint All-Domain Operations

• RCVs could directly support future Brigade Combat Teams commanders ability to see first, decide first, and make first contact with the enemy

• RCVs have the potential to reduce the risk to Soldiers, while simultaneously increasing the likelihood of mission success

Use or disclosure of data contained on the page is subject to restrictions on title page.

Recent RCV Accomplishments and Future Plans

• RCV Enterprise concluded the Phase I Soldier Operational Experiment (SOE) at Fort Carson, CO in August

• Awarded other transaction agreements for RCV(L) and RCV(M) surrogate prototypes to support Phase II SOE in 2022

• Continuing to develop and refine plans for additional technical risk reduction efforts in FY22 to FY24 in support of RCV programs of record

• Planning for RCV(L) program of record in FY23 and RCV(M) program of record in FY24

Use or disclosure of data contained on the page is subject to restrictions on title page.

Current RCV Acquisition Strategy

• Follow DoDI 5000.02 – ACAT I program

• Contract strategy to use integration vendor – integrate Control Station, Network, Platform, Software, and Payloads

• Separate competitive subcontract for RCV vehicle platforms – RCV(L) and RCV(M)

• Government furnished software and some payloads integrated through Modular

Open Systems Approach (MOSA) based architecture

•RCV(L) MS B in FY23 •RCV(L) FUE in FY28 •RCV(M) MS B in FY24 •RCV(M) FUE in FY30

FY20 FY21 FY22 FY23 FY24 FY25 FY26 FY27 FY28

Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4

O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S

Milestones - RCV(L)

RCV (L) Main Program Path

Milestones - RCV(M)

RCV (M) Main Program Path

DRFPRDP MS B MS C FRP

DP 2 DP 3 Acq KP #1 Acq KP #2 Acq KP#3 FUE IOC

AROC AROC

Initial Draft Program Plan

ITRA CDD ADM PRR

PCA OTRR FMR

Milestone B Docs

CDD Validated Final CDD

AoA

Contract Prep RFP Prop

Rec

Design/Integrate Award CDR

Build

TRR

Test (DT) LRIP LRIP Award

Full Rate Production

OT&E

LOG Development

DRFPRDP MS B MS C

AROC ADM AROC PCA

ITRACDD

Milestone B Docs

CDD Validated Final CDD

Contract Prep RFP Prop

Rec Award

Design/Integrate

CDR

Build

TRR

Test (DT) LRIP Award

LRIP

LOG Development

Use or disclosure of data contained on the page is subject to restrictions on title page.

RCVs will be Systems of Systems

Use or disclosure of data contained on the page is subject to restrictions on title page.

Government Furnished Equipment

• List is draft, likely not complete, and subject to change

• Anticipated Government Furnished Equipment

– Autonomy software

– Radios

– Warfighter Machine Interface software

– Aided Target Recognition software

– Lethality payloads (minus the turret)

• Potential Government Furnished Equipment

– Tethered unmanned aerial system

– Assured position, navigation, and timing

– Hostile fire detection

Use or disclosure of data contained on the page is subject to restrictions on title page.

My View on RCVs Biggest Challenge and Opportunity

• Autonomy and automation software are critical, perhaps the most critical, to effective operation of RCVs in contested environments

• Autonomy will not be ‘solved’ for combat vehicle applications in the near future and Soldiers will have to be “in-the-loop” for the foreseeable future

• For Soldiers operating / commanding RCVs, software development and maturation should primarily focus on two broad areas:

– Reducing the need for Soldier interaction with the system

– Improving the performance of the system when Soldiers must interact

Use or disclosure of data contained on the page is subject to restrictions on title page.

“Commonality” aka Where We Need Your Input

• We desire to maximize our return on investment (or minimize the amount of re-work) for sub-systems / components that should / could be common across classes of Robotic Combat Vehicles e.g., control stations and autonomy / automation software

• What’s the best business case for incentivizing innovation in regards to autonomy / automation software for generally military unique (or limited commercial interest) behaviors and functions?

Use or disclosure of data contained on the page is subject to restrictions on title page.

Questions?

Distribution Statement A – Approved for Public Release, Distribution Unlimited 49

Overview/Intent of RACER Phase 1 & BAA

Distribution Statement A – Approved for Public Release, Distribution Unlimited

DARPA TTO RACER Phase 1 - Intent and Funding Opportunity Overview

Stuart Young, PhD TTO Program Manager stuart.young@darpa.mil Office: (703) 526-2066 Mobile: (571) 302-2015

October 9, 2020

RACER

Distribution Statement A – Approved for Public Release, Distribution Unlimited

• RACER development plan

• RACER – unique insights

• RACER – developing autonomy algorithms and stacks

• RACER platforms overview and platform-developed autonomy

• RACER metrics overview

• RACER Government Furnished Equipment and Information (GFX)

• RACER DARPA-hosted Field Experiments overview

• Phase 1 BAA details

Agenda

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER Development Plan

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER – Unique Insights:

Improve Off-Road Speed Without Interventions

• Autonomy algorithms and stacks

• Implement real-time, online algorithm approaches that learn and auto-tune parameters on-the-fly

• Understand faster, think farther: succeed in autonomy at speed

• Transition algorithms/stacks to Government customers

• Off-road speed

• Demonstrate operationally relevant system speed on surrogates

• Speeds no longer limited by algorithms and processing

• Proof through performance demonstrated at DARPA-hosted field experiments

• Resiliency

• Algorithms/stacks proven across multiple terrain environments

• Generalize immediately to every environment

• Field based development

• Systematic develop-test-develop-test

• Many, many test cycles; many, many environments

• Enabled by Government Furnished Equipment and Information (GFX):

• Lowers program risk

• Maintains program focus on algorithm/stack development

• Establishes a level playing field and point of departure

Distribution Statement A – Approved for Public Release, Distribution Unlimited 54

Autonomous Mobility in Unstructured Off-Road Terrain:

RACER Insights to Developing Autonomy Algorithms and Stacks

Short Range Adequate 3D knowledge of the immediate vicinity around the platform

Medium Range 2D understanding of the world within sensor range but at the extents of or just beyond 3D range

Short

Medium

Long

Prior approach limits:

• LIDAR or stereo vision only to see openings and plan mid-range

Prior approach limits:

• Slow speed and brittleness from manually tuned parameters for obstacle classification, ground plane estimation, cost, planners, and vehicle dynamics

Global routes provided

Distribution Statement A – Approved for Public Release, Distribution Unlimited 55

Classic Vehicle Autonomy Stack

Detections, Correlation of Objects, Orientations to Current Vehicle

Tracks, Trajectories of Others

Long term Predictions, Paths/Plans/Intent of

Current Vehicle and Others

Motion Trajectories Of Current Vehicle

Steering, Speed, Acceleration to Current

Vehicle Controller

Sensors

Map Data

Sensor Data Detector Tracker Prediction Planning Control Actuation

Attribute Self Driving Vehicles Object Classes ~4 (car, truck, person, motorcycle/bike, etc.)

Accuracy cm Movement Environment Extremely dynamic obstacles Dimensionality Primarily 2D Map Dependence Fully dependent on high resolution data Mobility Dynamics Predictable, generally fixed Collision Tolerance 0.9999 – No forgiveness

Maps

Traditional Engineering Stack – On Road, Self Driving Vehicle From:

“A future with affordable self-driving vehicles”, R. Urtasun – Uber ATG, IEEE ICRA 2019 Keynote, 22 MAY 2019, Montreal, Canada

Distribution Statement A – Approved for Public Release, Distribution Unlimited

Attribute Self Driving Vehicles Object Classes ~4 (car, truck, person, motorcycle/bike, etc.)

Accuracy cm Movement Environment Extremely dynamic obstacles Dimensionality Primarily 2D Map Dependence Fully dependent on high resolution data Mobility Dynamics Predictable, generally fixed Collision Tolerance 0.9999 – No forgiveness

Attribute Self Driving Vehicles RACER Object Classes ~4 (car, truck, person, motorcycle/bike, etc.) 100s of classes (10s of just vegetation) Accuracy cm 10s of cm Movement Environment Extremely dynamic obstacles Generally static Dimensionality Primarily 2D Fully 3D due to slope Map Dependence Fully dependent on high resolution data Highly minimized, using coarse data Mobility Dynamics Predictable, generally fixed Complex, with varied types including slip Collision Tolerance 0.9999 – No forgiveness Tolerated – balance with risk Tactical Movement Tactical context of movement

Autonomy Stack Comparison – RACER Significantly Increases in Complexity

Traditional Engineering Stack – On Road, Self Driving Vehicle Adapted from:

“A future with affordable self-driving vehicles”, R. Urtasun – Uber ATG, IEEE ICRA 2019 Keynote, 22 MAY 2019, Montreal, Canada

Detections, Correlation of Objects, Orientations to Current Vehicle

Tracks, Trajectories of Others

Long term Predictions, Paths/Plans/Intent of

Current Vehicle and Others

Motion Trajectories Of Current Vehicle

Steering, Speed, Acceleration to Current

Vehicle Controller

Sensors Sensor Data Detector Tracker Prediction Planning Control Actuation

More Classes 3D Complex Dynamics

Dependent on SLAM-based State

Estimation

Map Data Maps Adversarial

Modeling & OpOrders

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER Autonomy Stacks? Insert Your Innovation

Map Data (minimized to none) Metric Planning & Execution

Simultaneous Localization and Mapping (SLAM)

Symbolic Planning & Execution Mission Planner Mission Planner

Global Planner Global Planner

Navigation ExecuterNavigation Executer

Mission Executer Mission Executer

BehaviorsBehaviors

(Localization in Environment)

Dynamic Vehicle Model

Detector Planning Control Actuation

Local Planner Local

Planner Local

Controller Local

Controller

Vehicle State EstimationVehicle State Estimation

Goal / Objective Function

Vehicle State

Mission Intent

Sensors Sensor Data

Traditional Engineering Stack

GFX RACER Baseline Stack

Sensors

Maps

Soldier (OpOrder)

Specific Vehicle Model

Cost Function

Innovation Approach?:

Inverse Reinforcement Learning

Innovation Approach?:

Reinforcement Learning

Innovation Approach?:

SLAM Off-Road, at Speed, over

Time/Distance

Perception

Innovation Approach?:

Deep Learning & Transfer Learning

Innovation Approach?:

Terrain Understanding &

Semantic Classifiers

Innovation Approach?:

End-to-End Learning

Distribution Statement A – Approved for Public Release, Distribution Unlimited 58

• Goal: establish traversal cost estimates for the environment

• Near-range perception system generates traversal cost estimates in proximity to the robot (blue region)

• These cost estimates are used to learn the mapping from difficult to interpret locale-specific features to traversal cost

• The resulting learned model can then be applied elsewhere to produce traversal cost predictions where no near-range perception estimates are available

• Course far range sensors (e.g. 100m LIDAR, wide-baseline stereo vision)

One of Many Existence Proofs of Real-Time, Online Algorithm Approaches Applicable to RACER DARPA UPI (2008) – Far-Range On-Line Learning (FROLL)

J.A. Bagnell, D. Bradley, D. Silver, B. Soffman, A. Stentz, “Learning for Autonomous Navigation – Advances in Machine Learning for Rough Terrain Mobility,” IEEE Robotics and Automation Magazine, Vol. 1070-9932, June, 2010, pp.74-84

D. Silver, J.A. Bagnell, A. Stentz, “Learning from Demonstration for Autonomous Navigation in Complex Unstructured Terrain,” International Journal of Robotics Research, Vol 29, Issue 12, June, 2010

Distribution Statement A – Approved for Public Release, Distribution Unlimited

Robot is “stuck” due to high grass

End-to-end learning allows robot to learn that it can drive through grass

One of Many Existence Proofs of Real-Time, Online Algorithm Approaches Applicable to RACER DARPA I20 Assured Autonomy (2020) – BADGR Self-Supervised Learning-Based Navigation

DARPA I2O Assured Autonomy Performer: G. Kahn, P. Abbeel, S. Levine, “BADGR: An Autonomous Self-Supervised Learning-Based Navigation System,”

Berkeley AI Research (BAIR), University of California, Berkeley, February 2020

On a subscale, low speed robot, UCB BADGR demonstrated:

• Current limits – LIDAR-only geometric reasoning is insufficient due to:

o Limited traversal through tall grass (obstacles) and avoidance of uneven terrain (poor slope estimation)

• BADGR insight – a fully automated, self-improving navigation system o End-to-end autonomous machine that can be trained with self-supervised data o Autonomously learned navigational affordances o Self-improves speed/paths in real time as it gathers more data o Generalizes to unseen environments

BADGR successfully reaches goal while avoiding collisions, while geometry-based policy is unable to make progress (falsely believes grass is un-traversable obstacles)

Generalizes to other off-road and urban environments

RACER to leverage auto-tuning and learning approaches/performers; will apply them to the full scale, off-road environment at speed problem

BADGR predicting which actions lead to uneven/bumpy terrain (left) or collisions (right)

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER Platforms Overview

Specification Phase 1 - LTATV Phase 2 - APD Configuration 4 wheel, Ackermann steering 6 wheel, zero-pivot skid-steer

Weight 1.3 T 9 T Overall Length 3.6 m 5 m Overall Width 1.5 m 2.5 m Mobility Power 42 kW 150 kW

Mobility Configuration Parallel hybrid-electric via electric shift CVT (anticipated)

Series hybrid-electric to 6 independent wheel stations

Distribution Statement A – Approved for Public Release, Distribution Unlimited 61

RACER Platform-Developed Autonomy (field evaluated)

RACER Phase 2

11DARPA Experiments & Evaluations

Teams Performers Propose & Develop:

• Autonomy stacks

• Short/medium-range autonomy approaches/algorithms for

A to B mobility

• Field site and field-ready team

Team Initial & Evolving

Autonomy Stacks

Evolving Coding Approaches/

Algorithm Modules

Local Test Site

Evolving Datasets

Diagnostic Tools

RACER Phase 1

22 33 44 55 66 77 88

Team Team TeamTeam

Constantly coding and driving at local test site to improve performance

Full integration, constantly coding;

driving long courses and global routes

• Objective: Performer teams develop off-road autonomy using a vehicle platform-based, agile develop-test-develop-test model that has been successful in the self-driving vehicle industry for rapid development

• Hypothesis: Autonomous off-road algorithms will be rapidly developed using existing development tools (including simulation) and platform/field-based agile develop-test-develop-test model.

Deliverables:

• Autonomy algorithm modules

Distribution Statement A – Approved for Public Release, Distribution Unlimited

Fast computation of vehicle dynamics and controls

Adaptive cost assessment and path planning

RACER Platform-Developed Autonomy - key technical challenges Short and medium range robotic sensing, path planning, and vehicle control

Adequate off-road sensing and fusion at speed in all environments

Scene interpretation, algorithm auto-tuning, and localization

Attribute RACER Need Object Classes 100s of classes (10s of just vegetation) Accuracy 10s of cm Sensing Range 20 0m Dimensionality Fully 3D due to slope Localization Vision-based, limited GPS, high shock/vibration Mobility Dynamics Complex, with varied types including slip Collision Tolerance Take advantage of all contact and encounters

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER Metrics

DARPA metrics to be achieved by this BAA

Metrics Phase 1 Phase 2

RACER

Complex terrain, long distance runs Average Autonomous Speed

5 m/s (18 km/h)

8 m/s (28.8 km/h)

Open terrain, short distance runs Autonomous Maximum Speed Runs

10 m/s (36 km/h)

20 m/s (72 km/h)

Interventions/km 0.5 0.1

Phase 2 Metrics

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER Government Team Activities and Contributions

• DARPA managed code repository/data sets

• GFX platform models, data sets, stacks

RACER Government Furnished Equipment and Information (GFX)

DARPA-hosted Field Experiments

RACER GFX FleetRACER Code Repository

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER Anticipated Government Furnished Equipment and Information (GFX) Overview

RACER GFX Fleet

Platforms Sensors Fleet Support

RACER Code Repository

RACER Development Environment

RACER

Data Store

Data Sets

Baseline Autonomy Stack

Distribution Statement A – Approved for Public Release, Distribution Unlimited 66

RACER Phase 1 anticipated GFX Platform – LTATV and sensors (notional/point of departure examples)

LIDAR: (e.g., Velodyne 32E, Ouster OS1/OS2); two in front, one in rear.

Stereo camera pairs (e.g.

Carnegie Robotics S21B); two pair in front and rear, one pair on either side.

GPU Processing Stack/Network

4-6 NVIDIA Drive AGX Orin Class GPUs or

Radar Arrays (e.g. NXP TEF810x) – multiple arrays

IMU (e.g., Honeywell

HG4930)

LTATV (Anticipated)

• hybrid electric

• drive-by-wire

• autonomous ready

4-6 Nuvo 7160GC GPUs or

Equivalent

* Requests to modifications or enhancements to GFX baseline to be considered. Modifications or enhancements, if approved, to be applied to GFX of all performers.

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER DARPA-hosted Field Experiments Overview

Tentative Locations Duration Schedule Phase 1

Experiment 1 Fort Irwin, CA 10 days 5 Months After Award (MAA)

Experiment 2 Yakima Training Center, WA 10 days 11 MAA

Experiment 3 Fort Irwin, CA 10 days 17 MAA

Phase 2 (notional)

Experiment 4 Fort Carson, CO 10 days 24 MAA

Experiment 5 Fort Drum, NY 10 days 30 MAA

Experiment 6 Yuma Proving Grounds, AZ 10 days 36 MAA

Experiment 7 Fort Benning, GA 10 days 42 MAA

Experiment 8 Fort Irwin, CA 10 days 48 MAA

Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Courses are in two course/performance classes:

• Complex terrain, long distance courses:

• ~5 km long or less (for Phase 1)

• Generally trail-less off-road natural terrain with vegetation, slope, discrete obstacles, and ground surface changes

• Intended to highlight non-stop autonomous movement/maneuver

• Open terrain, short distance courses:

• ~2.5 km long or less (for Phase 1)

• Generally intended to highlight speed performance of autonomy in off-road terrain

• Trails with less dense and fewer encounters of vegetation, slope, discrete obstacles, and ground surface changes

• Intended to highlight autonomy to the maximum extents of platform performance

• Courses will be made up of waypoints spaced 500 m - 1000 m apart.

• Multiple routes between waypoints will exist that can achieve RACER average speed metrics when driven by a human driver

• DARPA assesses program metrics from performer’s best run from both of the two course performance classes from DARPA-hosted field Experiment 3

• Overall Phase 1 performance will be based on the assessment of program metrics combined with evaluations of progress against the performer’s RACER Development and Demonstration Plan (RDDP) and the quality of program deliverables

DARPA-hosted Field Experiments - Courses

Distribution Statement A – Approved for Public Release, Distribution Unlimited

• For Phase 1:

• DARPA will specify goal points, waypoints, and course boundaries defined by GPS coordinates

• Proposer approaches may utilize non-WAAS GPS accurate to +/-10 m maximum

• Not guaranteed at all times in all environments

• Proposers approaches may utilize a pre-loaded 1:50,000 topological map

• No additional GPS or pre-loaded information may be used

• Performers may not rely on data or maps from prior runs

• Phase 1 metrics will be assessed using the LTATV platform

• Phase 2 metrics will be assessed using the LTATVs and the large-scale demonstration platform.

• Phase 2 complex terrain, long distance courses are planned to be 15-30 km or longer.

• Phase 2 open terrain, short distance courses are planned to be 4-5 km or longer.

DARPA-hosted Field Experiments – Courses (cont.)

Distribution Statement A – Approved for Public Release, Distribution Unlimited

For Phase 1:

• Experiments at a military training area/range

• DARPA gives:

• Waypoints

• Maneuver Corridor

• Courses selected knowing a human-driven platform can achieve RACER metrics on multiple routes between waypoints

• Off-road trails are a constant in training areas

• RACER complex terrain courses and route options will be designed to intermix with trails but not overly rely on them

• RACER complex terrain courses in the absence of trails will be the RACER standard

RACER DARPA-hosted Field Experiments Intent – Complex terrain, long distance courses

Waypoint

Waypoint

Maneuver Corridor

Distribution Statement A – Approved for Public Release, Distribution Unlimited

RACER DARPA-hosted Field Experiments Intent – Open terrain, short distance courses

Waypoint

Waypoint

Maneuver Corridor

Waypoint

WaypointWaypoint

Waypoint

For Phase 1:

• Experiments at a military training area/range

• DARPA gives:

• Waypoints

• Maneuver Corridor

• Courses selected knowing a human-driven platform can achieve RACER metrics on multiple routes between waypoints

• Off-road trails are a constant in training areas

• RACER open terrain courses and route options will be designed to intermix with trails

• RACER open terrain courses will prioritize speed on terrain while optimizing perception, planning, and control for such

Distribution Statement A – Approved for Public Release, Distribution Unlimited

• DARPA-hosted field experiment areas will constitute several square kilometers of natural maneuver terrain

• DARPA will establish a “RACER Cantonment Area” of base support for both the RACER Government Team and the performers

• It is anticipated that DARPA will provide operational infrastructure for a team of up to 20 on-site personnel per performer

• DARPA currently envisions the following general 10-day on-site experiment demonstration approach. Performers’ schedules should identify key events leading up to each field experiment or demonstration. Performers may recommend alternative test approaches for consideration.

• Five days for shakeout and system training in a general terrain “sandbox” facilitated by DARPA

• Five days for DARPA evaluation runs of a performer’s capability over multiple DARPA courses

• Evaluation infrastructure is currently envisioned to include:

• A boundary terrain corridor identified by GPS coordinates

• A global route, consisting of waypoints separated by 500-1,000 meters

• A DARPA safety vehicle with both a government and a performer rider/e-stop operator

• A wireless network for RACER connectivity across the entire test area

• A performer operator station

• A low bandwidth, standard set of telemetry messages to be exchanged between DARPA and performer platforms.

• For cost and efficiency purposes, the Government reserves the right to overlap performers during the DARPA-hosted field experiments

• Practice courses will be available for performer testing during the DARPA-hosted field experiments.

DARPA-hosted Field Experiments – Operations & General Expectations

Distribution Statement A – Approved for Public Release, Distribution Unlimited 73

BAA HR001121S0004 Details – RACER Phase 1

Distribution Statement A – Approved for Public Release, Distribution Unlimited 74

Questions:

• Please submit your written questions via the Zoom question feature

• This will allow the DARPA RACER team to capture the Question and Answer session for those unable to attend

• Answers to the questions will be provided later and posted to https://beta.sam.gov/

• Presentation materials may be made available after clearing public release

Proposers Day vs. BAA / Questions

Note:

In the event of a BAA posting, if there is any discrepancy between what is presented today and the BAA, the BAA takes precedence

Distribution Statement A – Approved for Public Release, Distribution Unlimited 75

Anticipated RACER BAA Dates

BAA HR001121S0004 – RACER Phase 1 BAA Posting Date: October 13, 2020 Abstract Due Date and Time: October 28, 2020, 4:00pm Eastern Time FAQ/Questions Due Date and Time: November 13, 2020, 4:00pm Eastern Time Proposal Due Date and Time: December 11, 2020, 4:00pm Eastern Time Contract Award: 3QFY21 (April to June 2021)

Distribution Statement A – Approved for Public Release, Distribution Unlimited

• Anticipated Awards: Up to 3

• Available resources depend on quality of proposals received

• Proposers looking for innovative, commercial-like contractual arrangements are encouraged to consider requesting Other Transactions

• Anticipated funds for Phase 1: $19.5M

• At this time, DARPA is soliciting full proposals for Phase 1 only.

• Phase 1 shall comprise:

• An 18-month Phase 1 base period

• Culminate in DARPA-hosted field experiment 3.

• A 3-month Phase 1 option.

• 3-month continuation effort that DARPA will exercise if needed to eliminate a gap in Phase 1 and Phase 2 performance

• Expected that the option period pace of activities and level of effort will be similar to the final 3 months of Phase 1

• Proposers shall also provide an initial program plan, initial schedule, and Rough Order of Magnitude (ROM) cost for a 30-month Phase 2 option

• DARPA plans to select 1 or more performers to continue on to Phase 2

• Selection of Phase 2 performers will be limited to Phase 1 prime contractors.

• Updated proposal guidance and a request for Phase 2 proposals will be provided to Phase 1 performers toward the end of Phase 1

In all cases, the Government contracting officer shall have sole discretion to select award instrument type, regardless of instrument type proposed, and to negotiate all instrument terms and conditions with selectees. DARPA will apply publication or other restrictions, as necessary, if it determines that the research resulting from the proposed effort will present a high likelihood of disclosing performance characteristics of military systems or manufacturing technologies that are unique and critical to defense. Any award resulting from such a determination will include a requirement for DARPA permission before publishing any information or results on the program.

BAA HR001121S0004 General Information

Distribution Statement A – Approved for Public Release, Distribution Unlimited 77

RACER - autonomy to drive off-road at speed What would an unmitigated success look like?

Drive autonomously:

• Fast enough to be operationally relevant

• Resilient enough to go anywhere

• Maneuver like a Soldier

Distribution Statement A – Approved for Public Release, Distribution Unlimited 78

ARL 6.1/6.2 Autonomy Stacks Dr. Ethan Stump ARL AI Maneuver/Mobility

Program Chief Scientist Dr. Jon Fink ARL AI Maneuver/Mobility

Mobility Lead

APPROVED FOR PUBLIC RELEASE

U.S. ARMY COMBAT CAPABILITIES

DEVELOPMENT COMMAND –

ARMY RESEARCH LABORATORY

Dr. Ethan Stump

Artificial Intelligence for Maneuver and Mobility: Program Chief Scientist

Computational and Information Sciences Directorate

09 OCT 2020

Government-Furnished Baseline Autonomy and Developer Tools for RACER

Dr. Jon Fink

Artificial Intelligence for Maneuver and Mobility: Mobility Lead

Computational and Information Sciences Directorate

• This presentation is intended to:

– Provide context for the baseline autonomy and explain its purpose in RACER

– Introduce the autonomy components and rationale

– Introduce the development tools being provided and rationale

• Outline

– Introduction and Background

– Baseline Autonomy Stack Discussion

• (1) Perception, (2) Localization and Mapping, (3) Navigation, (4) Symbolic Planning and Execution, (5) Simulation

– RACER Development Environment

• Tools, Example Workflow

– Summary and Conclusion

PURPOSE AND OUTLINE

• Strengthen program by supporting two key audiences:

– Academia: Integrated starting point can lower risk, allow more focused development, performance comparisons

– Industry: Framework to understand government’s perspective, demonstrate value, communicate transitions

• Army has invested a lot in R&D with academia and industry over the past 25+ years

– Baseline is a snapshot of some of this collaborative development

– Provided in a way that people can take advantage of it for this program, if they wish

• Challenge Programmatic Norms

– Army modernization strategy calls for less requirements engineering, more iterative development

– Progress always measured against a working system

WHY PROVIDE BASELINE AUTONOMY?

• ARL has advanced numerous concepts of military use of small unmanned systems and robot teaming

– Programs: MAST, RCTA, DCIST

– Locations: Off-road, urban, tunnels

• Outcomes

– Semantic segmentation (HIM)

– Navigation planning (SBPL)

– SLAM (GTSAM, Omnimapper)

– Language grounding (H2SL)

• This baseline represents work from many academic and industry collaborations

– Driven by ARL research consortiums

BACKGROUND

Stay to the right of the car; screen the back of the building that is behind the car.

F(2

ES1

• ARL is the US Army’s corporate laboratory

– Performs the majority of Army basic and applied research

– Has pioneered use of consortiums and open campus to bring academic and industry community together to solve Army challenges

• ARL drives research transition into larger Army S&T community

– Autonomy stack being used as an early entry point into the development pipeline that feeds GVSC, the NGCV CFT, and the PEOs

• ARL has a long history in vehicle autonomy

– Tapped pathfinding work from Europe in late 90s (Ernst Dickmanns)

– Demo III laid groundwork for first DARPA Grand Challenge

– Gen-1 LIDAR was outgrowth of first ARL Robotics Consortium

– Focus shifted to small vehicles as self-driving auto industry started

BACKGROUND

• ROS1 (full distribution ~300 packages)

– Mainly C++, some python

• Monolithic git repository

• Mostly an accumulation of internal, MAST, RCTA capabilities

• Core functions

– Perception (lidar and vision)

– Localization and mapping

– Metric planning and execution

– Symbolic planning and execution

– Simulation

• Some features may not be relevant for

RACER

– But who knows…

• Intended platform set:

OVERVIEW OF BASELINE AUTONOMY STACK

Perception Pipeline

SLAM

Metric Planning & Execution

Symbolic Planning & ExecutionNatural

Language Understanding

Behaviors

Object detection

(bbox) Tracking Position

Estimates

Cache and Fuse

Observations (Static Objects)

Pixel Classification

Project Terrain and Cache

Classify Points

Project Grids and Cache Render Grid

Render Terrain

Render Cloud

SLAM

Backend

(GTSAM)

Cache and Perform ICP

Odometry

LIDAR

RGB

Wheel speeds

IMU

Global Planner Local Planner Local Controller

Navigation Executor

Mission Planner

Mission Executor

Go to Object Behavior

Follow Behavior

Human detection

(mask)

Object Detection/ Tracking

Data Association and Tracking

(Dynamic objects)

Pose Estimates cmd_vel

Metric Goals:

* Go to region (x, y, th, R)

* External path ([(x, y, th, R)]

Symbolic Goals

* Go to object

* Go behind object

* Follow object objects

Depth

GPS

pose graph (map->odom corrections)

EXAMPLE MAPPING OUTPUTS

GVSC

MRZR

ARL

Warthog

• Role

– Feature extraction from raw sensor data

• Perception Concept

– Collection of ROS nodes/nodelets for transforming common sensor messages

– Run standalone or embed in localization/mapping pipeline

• Basic geometric perception

– Traditional point cloud classification for traversability

– 2.5D assumption and projection into occupancy grids

– Costs determined from heuristics like exponential inflation

• Semantic segmentation of images and point clouds

– ROS wrappers for FCNN, ICNet, RandLA-Net

– Projection onto terrain to provide semantic features for learned costmaps

– Projection onto LIDAR data to improve classification

• Primary Gaps

– Focused on 2D model of the world (occupancy grid generation system)

PERCEPTION CAPABILITIES

Zhao, H., Qi, X., Shen, X.,…