Robust and Reactive Decision-making and AI Planning of Collaborative and Agile Robots in Complex Environments

funded by National Science Foundation IIS, 2019-2022

A core research direction of The Laboratory for Intelligent Decision and Autonomous Robots (LIDAR) at Georgia Tech focuses on formal methods and decision-making algorithms of dynamic terrestrial locomotion and aerial manipulation in complex and human-surrounded environments. We aim at scalable planning and decision algorithms enabling heterogeneous robot teammates to dynamically interact with unstructured environments and collaborate with humans. In particular, we are interested in robust task and motion planning approaches that (i) abstract and unify diverse, complex low-level robot dynamics generally possessing under-actuated, hybrid, and nonlinear features; (ii) computationally efficient, safe and reactive decision-making algorithms explicitly taking into account dynamic environmental events and human motions. One of our long-term goals is to achieve a hierarchical and scalable planning framework with the following objectives: (i) robust, non-periodic motion planners and control barrier certificates for versatile terrestrial and aerial maneuvering; (ii) game-based reactive task planner in response to diverse and possibly adversarial environmental events; (iii) novel multi-agent decision-making approaches that decompose the entire robot team into multiple sub-teams with receding horizon approaches. We will adopt algorithmic methods at the interaction of formal methods, multi-agent systems, robust control, and machine learning. Formal guarantees such as robustness, safety, and runtime performance will be targeted. The experimental performance will be evaluated on the Buzzy Cassie robot in the lab. Please stay tuned for more updates.

Risk-sensitive Contact-implicit Trajectory Optimization and Stabilization

supported by Draper Laboratory and Toyota Research Institute.

Contact-implicit trajectory optimization algorithms are a powerful set of tools for designing dynamic robot motions involving physical interaction with the environment. However, the trajectories output by these algorithms can be difficult to execute in practice due to several common sources of uncertainty: robot model errors, external disturbances, and imperfect surface geometry and friction estimates. To account for these uncertainties in a robust optimization framework, one typically aims to design a cost function that includes some measure of closed-loop performance variance. Unfortunately, the discontinuous nature of contact creates numerical challenges (e.g., non-differentiable cost functions, rapidly diverging trajectories) that must be addressed. We propose a modification to existing direct contact-implicit trajectory optimization methods that employs a sampling scheme inspired by the unscented transform to compute a cost function that computes a local measure of trajectory variance. To ensure the differentiability of the resulting cost function, we use employ a convex approximation to the linear complementarity formulation of rigid-body dynamics. The algorithm is validated in simulation and hardware manipulation experiments.

Reactive Task and Motion Planning for Robust Whole-Body Locomotion in Constrained Environments

Contact-based decision and planning methods are increasingly being sought for task execution in humanoid robots. However, formal methods from the verification and synthesis communities have not been yet incorporated into the motion planning sequence for complex mobility behaviors in humanoid robots.

This study takes a step toward formally synthesizing high-level reactive planners for unified legged and armed locomotion in constrained environments. We formulate a two-player temporal logic game between the contact planner and its possibly adversarial environment. The resulting discrete planner satisfies the given task specifications expressed in a fragment of temporal logic. The resulting commands are executed by a low-level 3D phase-space planner algorithm. We devise various low-level locomotion modes based on centroidal momentum dynamics. Provable correctness of the low-level execution of the synthesized discrete planner is guaranteed through the so-called simulation relations. Simulations of dynamic locomotion in constrained environments support the effectiveness of the hierarchical planner protocol.

The proposed whole-body locomotion in constrained environments could be an ideal scenario to the newly announced DARPA Subterranean Challenge 2018-2021 for underground operation augmentation.

Collaboration: Ufuk Topcu (Aerospace Engineering, UT Austin) and Luis Sentis (Head of HCR Lab, Aerospace Engineering, UT Austin).

Related publications:

Y. Zhao, Y. Li, L. Sentis, U. Topcu, J. Liu. Reactive Task and Motion Planning for Robust Whole-Body Dynamic Locomotion in Constrained Environments. International Journal of Robotics Research, Under Review, 2018. [preprint]

Ye Zhao, Ufuk Topcu and Luis Sentis. High-Level Planner Synthesis for Whole-Body Locomotion in Unstructured Environments. IEEE Conference on Decision and Control (CDC), 2016. [video] [Supplementary Material]

Ye Zhao, Ufuk Topcu and Luis Sentis. Towards Formal Planner Synthesis for Unified Legged and Armed Locomotion in Constrained Environments. Proceedings of Dynamic Walking, Oral Presentation. 2016.

Robust Optimal Phase Space Planning of Agile Bipedal Locomotion over Diverse Terrain Topologies

supported by Office of Naval Research (ONR), [grant #N000141210663].

This line of research presents a theoretical framework for planning and control of agile bipedal locomotion based on robustly tracking a set of non-periodic keyframe states. Formulated a hybrid phase-space planning and control framework which includes the following key components:

(1) a step transition solver that enables dynamically tracking non-periodic apex or keyframe states over various types of terrains,

(2) a robust hybrid automaton to effectively formulate planning and control algorithms,

(3) a phase-space metric to measure the distance to the planned locomotion manifolds, and

(4) a hybrid control method based on the previous distance metric to produce robust dynamic locomotion under external disturbances.

Compared to other locomotion frameworks, we have a large focus on non-periodic gait generation and robustness metrics to deal with disturbances.

Collaboration: Benito Fernandez (Mechanical Engineering, UT Austin) and Luis Sentis.

Related publications:

Ye Zhao, Benito Fernandez and Luis Sentis. Robust Optimal Planning and Control of Non-Periodic Bipedal Walking With A Centroidal Momentum Model. The International Journal of Robotics Research. 36(11): 1211-1243. September, 2017. [video] [ArXiv] [code] [online]

Ye Zhao, Benito Fernandez and Luis Sentis. Robust Phase-Space Planning for Agile Legged Locomotion over Various Terrain Topologies. Proceedings of Robotics: Science and Systems (RSS), 2016. [video] [slides] [Acceptance Rate: 20.6%]

Ye Zhao, Luis Sentis. A Three Dimensional Foot Placement Planner for Locomotion in Very Rough Terrains. IEEE-RAS International Conference on Humanoid Robots (Humanoids), 2012. [Google Citation: 44]

Exploring Visually Guided Human Locomotion over Rough Terrain and Extreme Environment

This line of research implements impedance control on our series elastic actuators and Hume bipedal robot. We formulated a novel Operational Space Control by incorporating series elastic actuator dynamics and feedback delays. Maximized Cartesian impedance range capability is achieved for both compliant interaction and accurate Cartesian tracking.

Collaboration: Nicholas Paine (CTO of Apptronik), Donghyun Kim (HCR Lab, UT Austin), Gray Thomas (HCR Lab, UT Austin) and Luis Sentis.

Related publications:

Ye Zhao, Nicholas Paine, Steven Jorgensen, and Luis Sentis. Impedance Control and Performance Measure of Series Elastic Actuators. IEEE Transactions on Industrial Electronics. In Press. 2017. [video] [code]

Donghyun Kim, Ye Zhao, Gray Thomas, Benito Fernandez and Luis Sentis. Stabilizing Series-Elastic Point-Foot Bipeds using Whole-Body Operational Space Control. IEEE Transactions on Robotics. 2016, 32(6), 1362-1379. [video] [ArXiv] [Best WBC paper award finalist] [Best WBC video award finalist]

Ye Zhao, Nicholas Paine and Luis Sentis. Feedback Parameter Selection for Impedance Control of Series Elastic Actuators. IEEE-RAS International Conference on Humanoid Robots (Humanoids), 2014. [Google Citation: 11]

Ye Zhao and Luis Sentis. Passivity of Time-Delayed Whole-Body Operational Space Control with Series Elastic Actuation. IEEE-RAS Conference on Humanoid Robots (Humanoids), 2016. [Supplementary Material]

Stability and Impedance Performance Limits of Distributed Feedback Controllers with Latencies

supported by NASA Johnson Space Center, NSF/NASA NRI Grant [grant #NNX12AM03G].

We implemented a distributed high impedance control on our UT linear actuator and omnidirectional mobile robot Trikey. Effects of feedback delays and filtering are taken into consideration. Analyzed physical reasons for phase margin sensitivity discrepancy to different feedback delays. A servo breakdown rule for distributed control architecture implementation is proposed.

Collaboration: Nicholas Paine (CTO of Apptronik), Kwan Suk Kim (HCR Lab, UT Austin) and Luis Sentis.

Related publications:

Ye Zhao, Nicholas Paine, Kwan Suk Kim and Luis Sentis. Stability and Performance Limits of Latency-Prone Distributed Feedback Controllers. IEEE Transactions on Industrial Electronics. 2015, 62(11), 7151-7162. [video1] [video2] [ArXiv]

[Journal Impact Factor: 6.498, H5-index: 111]

Ye Zhao, Nicholas Paine and Luis Sentis. Sensitivity Comparison To Loop Latencies Between Damping Versus Stiffness Feedback Control Action In Distributed Controllers. ASME 2014 Dynamic Systems and Control Conference (DSCC), 2014.

Active Suspension Control in Finite Frequency Domain

funded by the National Natural Science Foundation (NSF) of China [grant#60825303].

This line of research studied H-infinity control problem for active seat suspension systems and designed a dynamic output feedback controller of order equal to the plant. Considered actuator input delay, actuator output force limitation and controlled output constraints. Utilized finite frequency approach to achieve better disturbance attenuation performance for Human-sensitive frequency range.

Collaboration: Weichao Sun and Huijun Gao (Harbin Institute of Technology).

Figure. Quarter-car model with an active suspension Figure. The seat-driver model of 3 degree-of-freedom

Related publications:

Weichao Sun, Ye Zhao, Jinfu Li and Huijun Gao. Active Suspension Control with Frequency Band Constraints and Actuator Input Delay. IEEE Transactions on Industrial Electronics. 2012, 59(1): 530-537. [Google Citation: 155, Journal

Impact Factor: 6.498, H5-index: 111] [Thomson Reuters "Highly Cited Paper", top 1% citation in Engineering Category]

Weichao Sun, Jinfu Li, Ye Zhao and Huijun Gao. Vibration Control for Active Seat Suspension Systems via Dynamic Output Feedback with Limited Frequency Characteristic. Mechatronics. 2011, 21(1): 250-260. [Google Citation: 62, Journal Impact Factor: 1.726, H5-index: 35]

Platform Design of quarter car active vehicle suspension

supported by the National Innovative Project of HIT.

Duration: 2009.06 – 2010.02

Details: Designed a mechanical system of the hydraulic actuated quarter-car suspension.

Collaboration: Jinfu Li, Xincheng Ma and Huijun Gao (Harbin Institute of Technology).

Analysis and Synthesis of deterministic and stochastic switched systems

funded by the National Natural Science Foundation of China [grant#60904001].

This line of research analyzed the stability of Neural Networks with Markov jumping parameters and proposed a more practical scenario of the statistics of Markov mode transitions.

Collaboration: Lixian Zhang (Harbin Institute of Technology) and Shen Shen (MIT EECS).

Figure. Discrete-time Markov transition probability matrix

Related publications:

Ye Zhao, Lixian Zhang, Shen Shen and Huijun Gao. Robust Stability Criterion for Discrete-time Uncertain Markovian Jumping Neural Networks with Defective Statistics of Modes Transitions. IEEE Transactions on Neural Networks. 2011, 22(1): 164-170. [Google Citation: 67, Journal Impact Factor: 2.633, H5-index: 49]