Description:
OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy, Integrated Network System of Systems
OBJECTIVE: An autonomous legged robotic control system capable of navigating highly uneven, obstructed, and uncertain terrain.
DESCRIPTION: The future Warfighter will require autonomous robotic systems to traverse highly uneven, obstructed, and uncertain terrain at speed. Legged platforms are clear frontrunners to meet this requirement, but the control of such systems presents a substantial engineering challenge. However, recent developments in hybrid dynamical systems (the branch of control engineering science that effectively models legged systems) and computational capability suggest that the time to address this challenge has arrived. New techniques in signal filtration and uncertainty characterization may be refined to create a controller capable of guiding a robotic platform across terrain that, up until now, has been impassable by an autonomous agent. Successful performers will have to prove the validity of novel physics-based models and control frameworks for a quadruped robot in question for wide arrays of tasks and demonstrate superiority of this paradigm over learning-based control in specific situations. The results will be further streamlined and tested on current quadruped robots.
PHASE I: Design, develop, and validate improved techniques for state estimation and uncertainty propagation in model predictive control of hybrid dynamical systems - specifically quadruped robots in dynamic and uncertain environments. Demonstrate proof-of-concept of this new control paradigm, and quantify its efficacy over the current state-of-the-art. This demonstration should illustrate the ability of a quadruped robot to successfully autonomously navigate a test environment featuring sharply uneven terrain (roots and rocks whose characteristic length are on the order of, and slightly larger than, that of the quadruped foot) hidden underneath grass or grass-like obstructions whose height is on the order of the robot’s. A successful demonstration will permit a quadruped to traverse ten body lengths at 0.5 body lengths per second over flat but uneven terrain featuring ground level variance and grass-like obstructions not exceeding 20% of the robot's height.
PHASE II: Design, develop, and validate broad techniques for state estimation and uncertainty propagation across a wide array of physical environments in which a quadruped robot may operate. Demonstrate integration with existing novel perception and sensing capability in a path-planning exercise whose terrain includes obstructions like those in the demonstration of Phase I. Phase II should extend the methodologies of proprioception developed in Phase I to enable increased performance. Compare the efficacy of this new controller against that of traditional techniques such as deep reinforcement learning (DRL) controllers or Model Predictive Control (MPC). A successful demonstration will permit a quadruped to traverse a five body-length incline of +/- 20 degrees with root-like obstructions and slippery surfaces at 0.3 body lengths per second.
PHASE III DUAL USE APPLICATIONS: The end-state control architecture should be mature enough to extrapolate locomotor performance to any number of scenarios, environments, and robotic platforms. The ideal resulting controllers will feature selective frameworks (such as a framework that could choose between MPC, DRL, etc.), and the inherent ability to determine what control technique is most effective for the task at hand. Production-ready controllers will also enable a robotic platform to extract itself from a ”stuck” position in brush, soft soil, and/or rocky terrain.
REFERENCES:
1. Christopher Allred, Mason Russell, Mario Harper, and Jason Pusey. Improving methods for multi- terrain classification beyond visual perception. In 2021 Fifth IEEE International Conference on Robotic Computing (IRC), pages 96–99. IEEE, 2021.
2. Berk Altın and Ricardo G Sanfelice. Model predictive control for hybrid dynamical systems: Sufficient conditions for asymptotic stability with persistent flows or jumps. In 2020 American Control Conference (ACC), pages 1791–1796. IEEE, 2020.
3. Taylor Apgar, Patrick Clary, Kevin Green, Alan Fern, and Jonathan W Hurst. Fast online trajectory optimization for the bipedal robot cassie. In Robotics: Science and Systems, volume 101, page 14, 2018.
4. Max Austin, John Nicholson, Jason White, Sean Gart, Ashley Chase, Jason Pusey, Christian Hubicki, and Jonathan E Clark. Optimizing dynamic legged locomotion in mixed, resistive media. In 2022 IEEE/ASME International Conference on Advanced Intelligent Mechatronics (AIM), pages 1482–1488. IEEE, 2022.
5. Yanran Ding, Abhishek Pandala, Chuanzheng Li, Young-Ha Shin, and Hae-Won Park. Representation- free model predictive control for dynamic motions in quadrupeds. IEEE Transactions on Robotics, 37(4):1154–1171, 2021.
6. Wei Gao, Charles Young, John Nicholson, Christian Hubicki, and Jonathan Clark. Fast, versatile, and open-loop stable running behaviors with proprioceptive-only sensing using model-based optimization. In 2020 IEEE International Conference on Robotics and Automation (ICRA), pages 483–489. IEEE, 2020.
7. Philip Holmes, Robert J Full, Dan Koditschek, and John Guckenheimer. The dynamics of legged locomotion: Models, analyses, and challenges. SIAM review, 48(2):207–304, 2006.
8. Donghyun Kim, Jared Di Carlo, Benjamin Katz, Gerardo Bledt, and Sangbae Kim. Highly dynamic quadruped locomotion via whole-body impulse control and model predictive control. arXiv preprint arXiv:1909.06586, 2019.
9. Daniel E Koditschek. What is robotics? why do we need it and how can we get it? Annual Review of Control, Robotics, and Autonomous Systems, 4:1–33, 2021.
10. Nathan J Kong, J Joe Payne, George Council, and Aaron M Johnson. The salted kalman filter: Kalman filtering on hybrid dynamical systems. Automatica, 131:109752, 2021.
11. Matthew D Kvalheim, Paul Gustafson, and Daniel E Koditschek. Conley’s fundamental theorem for a class of hybrid systems. SIAM Journal on Applied Dynamical Systems, 20(2):784–825, 2021.
12. Hai Lin and Panos J Antsaklis. Modeling of hybrid systems. In Hybrid Dynamical Systems, pages 11–64. Springer, 2022.
13. Yuan Lin, John McPhee, and Nasser L Azad. Comparison of deep reinforcement learning and model predictive control for adaptive cruise control. IEEE Transactions on Intelligent Vehicles, 6(2):221–231, 2020.
14. David M Mackie. Analysis of army doctrine documents with respect to artificial muscle and robotic mules. Technical report, CCDC Army Research Laboratory, 2020.
15. Takahiro Miki, Joonho Lee, Jemin Hwangbo, Lorenz Wellhausen, Vladlen Koltun, and Marco Hut- ter. Learning robust perceptive locomotion for quadrupedal robots in the wild. Science Robotics, 7(62):eabk2822, 2022.
16. Marion Sobotka. Hybrid dynamical system methods for legged robot locomotion with variable ground contact. PhD thesis, Technische Universit¨at Mu¨nchen, 2007.
17. Arjan J Van Der Schaft and Hans Schumacher. An introduction to hybrid dynamical systems, volume 251. springer, 2007.
KEYWORDS: Robotics, Control, Dynamical Systems, Hybrid Dynamical Systems, Model Predictive Control, Perception, Proprioception, Exteroception, Path Planning, Nonlinear Systems
