Congratulations (Dr.) Jiayi Wang for a successful PhD Defence
Title: Online Receding Horizon Planning of Multi-Contact Locomotion
Congratulations to Jiayi Wang for successfully passing his PhD viva on 5th June 2023. His thesis is entitled 'Online Receding Horizon Planning of Multi-Contact Locomotion'.
Short lay summary:
Legged robots such as humanoids and quadrupeds can overcome terrain irregularities by using multiple contacts between their limbs and the environment. Such capability makes them particularly useful for tasks taking place in unstructured environments, e.g., exploring disaster zones or inspecting construction sites that are hazardous and dangerous for humans.
Nevertheless, achieving reliable legged locomotion in the real world is still a challenging task. One of the issues is that when operating in the real world, legged robots can encounter unexpected disturbances, such as dynamic changes in the environment or state deviations induced by external force perturbations. These unexpected events can cause the pre-planned motion to become invalid, and continuing to execute the current motion will cause the robot to fall.
To enable reliable operation in response to unexpected disturbances, legged robots necessarily require the capability to online (re)-plan their motions. Unfortunately, due to the combinatorial nature of the contact planning and the high-dimensionality and non-convexity of the robot dynamics, computing multi-contact locomotion plans is often time-consuming.
In this thesis, we develop novel methods to accelerate the computation speed of multi-contact locomotion planning. The core idea of our methods is to find simplifications to the original planning problem. To this end, we propose to introduce model simplifications along the planning horizon. Furthermore, we also explore the usage of machine learning techniques to encode domain knowledge that can be used to bootstrap the computation. We evaluate the computation performance of our methods through rigorous simulation studies. Owing to the computation advantage of our methods, we demonstrate online multi-contact locomotion (re)-planning in real-world robot experiments.
Jiayi's examiners were:
Dr Ludovic Righetti (New York University, External)
Dr Michael Herrmann (University of Edinburgh, Internal)