Research

Agile Quadruped Jumping Through Full Centroidal Momentum Model Predictive Control

Current progress on controlling quadruped robots through dynamic jumping and running motions. I designed a convex MPC controller utilizing full centroidal dynamics and centroidal momentum matrix. The designed controller is able to regenerate reference trajectory in real time, and robust enough to be directly deployed on hardware without low-level whole-body controller. Preliminary result for continuous 90 degrees jump turn is shown here on the MIT mini-cheetah platform. This work is accepted to appear in IEEE Robotics and Automation Letters (RA-L) later this year. Pre-print can be found here.

Momentum-Aware Planning Synthesis for Dynamic Legged Locomotion

In this project, we developed a bi-level trajectory optimization framework for generating dynamic and agile motion for quadruped robots. Our framework consists of two optimization components: a centroidal optimization using reduced order centroidal dynamics with momentum as decision variables; and a DDP-based full dynamics optimization. Additional consensus constraints are enforced to bridge the two components. Here are sample trajectories, for quadruped A1, generated using developed framework. A journal extension of this project is currently in progress.

Extending Riemmanian Motion Policies to Underactuated Wheeled-Inverted-Pendulum Robots

For this project, we developed a reactive motion planning pipeline for wheeled-inverted-pendulum based humanoids. We proposed a dynamics decomposition technique that splits the full dynamics of the robot into fully actuated and underactuated subsystems. Leveraging Riemannian motion policies, differential dynamic programming, we were able to independently plan trajectories for fully actuated and underactuated dynamics, which are then combined into a single motion plan through null-space projection. This work is published at ICRA 2020

Adaptively Robust Control Policy Synthesis Through Riemannian Motion Policies

In this project, we proposed a robust control Lyapunov function based tracking controller, capable of handling large magnitude of dynamics uncertainties and external disturbances. Additionally, we devised an automatic robust gain adaptation procedure that guarantees the stability and boundedness of the proposed controller. Experimental result on KUKA iiwa platform is shown here, for object tracking and obstacle avoidance. This work is published at the May issue of Control Systems Letters 2021

Distributed Optimal Control Framework based on Coordinate Descent Optimization for Multi-Agent Robots

In this paper, we present a distributed optimal control framework for a multi-agent robotics system based on coordinate descent optimization. Our framework exploits the underlying graph topology to compute the optimal control trajectory in a distributed manner. To show the efficacy of the framework, we apply it to a problem where a team of robots is tasked with establishing a communication link between source and destination while minimizing the overall system mobility and communication energy cost. Here is a sample of our algorithm running on the Gatech Robotarium platform using a cluster of 8 mobile robots. This work is published at CASE 2021