How to make a robot walk in challenging environments? Insights from fifteen years of research on multi-contact planning

by Steve Tonneau (University of Edinburgh), Victor Levé-Lin (University of Edinburgh)

Thursday 29 Jan 2026, 14:30 → 15:30 Europe/Zurich

774/R-013 (CERN)

Talk 1: Steve Tonneau 

 

Title: How to make a robot walk in challenging environments? Insights from fifteen years of research on multi-contact planning

 

Abstract:

In this talk, I discuss how legged robots can be made to walk and interact with their environment in challenging conditions. I draw on insights from fifteen years of research on multi-contact planning, and relate contributions from both the robotics and computer graphics communities, highlighting how these fields have often addressed similar problems from different perspectives.

I review the respective strengths and limitations of sampling-based and dynamic programming approaches, and compare them with methods commonly studied in artificial intelligence. Finally, I present experimental results on real robots and discuss the practical difficulties encountered when moving from theory to practice, with perception remaining a central challenge.

 

Bio: Steve Tonneau is a reader (Associate Professor) at the University of Edinburgh. He defended his Phd in 2015 after 3 years in the INRIA/IRISA Mimetic research team, and pursued a post-doc in robotics at LAAS-CNRS in Toulouse, within the Gepetto team. His research focuses on motion planning for legged robots, with a specific interest for combinatorics. Applications include computer graphics animation and robotics.

---

Talk 2: Victor Levé-Lin:

Title: Scaling Whole-body Multi-contact Manipulation with Contact Optimization

Abstract: We consider the issue of providing humanoid robots with the ability similar to humans to autonomously perform whole-body manipulation tasks. In this context, the infinite possibilities for where and how contact can occur on the robot and object surfaces hinder the scalability of existing planning methods, which predominantly rely on discrete sampling. Given the continuous nature of contact surfaces, gradient-based optimization offers a more suitable approach for finding solutions. Our work addresses the problem of suitable representations of the robot and object surfaces for continuous optimization, and cost functions designs to efficiently guide the manipulation planning. We demonstrate through experiments that our method improves drastically the convergence of planning over the state of the art and is accurate enough to be implemented on hardware.

Bio: Victor Levé is a third year Ph.D. student at the University of Edinburgh, with his research aiming at providing Autonomous Whole-body Multi-contact Manipulation skills to humanoid robots. 
Before his Ph.D., he worked for about ten years on the engineering design and integration of humanoid robots for the industry at Kawada Robotics (https://www.kawadarobot.co.jp/en/) in Japan.