Research

A full publication list, sortable by topic area, is available here.

Legged Controls

The Legged Controls Group focuses on unlocking the potential of legged platforms to navigate unstructured environments. We promote agile and robust locomotion through computationally efficient algorithms that allow the robot to quickly reason about its abilities and surroundings. We focus on all aspects of this challenge, from high-level planning to low-level control, perception and mapping to state estimation and learning. Some of our recent works include: fast global motion planning, MPC for legs and tails, and trajectory optimization.

Hybrid System Modeling, Estimation, and Control

This project seeks to further our understanding of how states evolve in hybrid systems to improve state estimation and control algorithms. Much of the recent work in this direction has focused on the use of saltation matrices, which describe the instantaneous uncertainty updates which occur during hybrid events such as leg touchdown. Some of the associated publications for this research direction include: Kalman filtering, contact localization, iLQR, and hybrid event shaping.

Learning to Drive Off-Road

Many real world tasks require robots to navigate through unstructured outdoor environments. These environments often include rough terrain with unavoidable obstacles, such as rocks, and unknown physical properties, such as contact friction. Autonomous navigation through such environments is difficult as the robot experiences complex, non-static, and divergent robot dynamics. We aim to enable off-road driving by having the robot learn from previous driving experience in real world and simulated environments to gain a probabilistic understanding of its dynamics in new environments. Given this probabilistic understanding, the robot can then perform robust decision making to find safe routes over difficult terrain to its destination.

Terramechanics and Environmental Monitoring

We want robots that can not just move over terrain but actually interact with the environment to enable planetary exploration and environmental monitoring applications. To achieve this, we work to develop new physical models of how the robot interacts with terrain features like soil and underbrush, new behaviors that leverage these robot-terrain interactions, and full systems that can go out and robustly work in natural environments. For example, nonprehensile terrain manipulation seeks to leverage incidental robot-terrain interactions for intentional manipulation of the environment, such as digging trenches with a rover's wheel.

Robotic Climbing

Climbing robots can operate in steep and unstructured environments that other ground robots cannot access. Robots like LORIS and T-RHex use compliant arrays of small hooks called microspines to scale rocky surfaces, enabling scientific exploration of cliff faces and subterranean caves. Similarly, magnetic adhesion enables wall-climbing robots like Sally to automate the inspection of steel structures such as bridges and storage tanks. The development of these robots involves aspects ranging from mechanism design and system integration to low-level force control and high-level path planning.