Rock Climbing Robots Workshop 2026

About the Workshop

Despite rapid progress in humanoid robotics, complex multi-contact motion planning for highly dynamic tasks remains a fundamental challenge. This workshop explores the intersection of generative AI (diffusion models & transformers) for high-level contact planning and reinforcement learning for whole-body control, using rock climbing as a challenging benchmark.

Contact Planning Whole-Body Control Diffusion Models Reinforcement Learning Humanoid Climbing Multi-Contact Robotics Legged Locomotion Generative AI

Curriculum Overview

Climbing Demo

Hackathon

The goal of the Hackathon is to climb small walls using humanoid-shaped robots.

The three challenges below can be tackled independently — teams are free to focus on any single challenge or combine several of them.

Challenge 1 — Perception

Build a 3D model of the climbing wall and estimate the robot's pose in real time. Can be addressed independently of planning and control.

Challenge 2 — Planning

Generate a feasible path and contact sequence for climbing the wall. Fiducial markers can be used so perception is not required.

Challenge 3 — Control

Whole-body execution of the planned motion on a physical robot platform. A pre-computed contact plan can be provided so planning is not required.

Climbing robot prototype developed by the organizers.

Practical Information

A simulator will be released by the organizers (date T.B.D.)
A small-size robotic platform will be provided
Fiducial markers available — teams can skip perception and focus on planning & control
Official rules are currently being written

Microban in action

Hardware — in progress

Hardware specifications

Microban will be used as a base to design a climber for the Hackathon.

Specifications for the small climbing wall and the robotic platforms are still being designed — they will be published here as soon as they are finalized.

Rhoban/microban on GitHub

Call for Participation

We welcome all students and researchers interested in robotics and/or AI, regardless of experience level.

Who should attend?

Undergraduate & graduate students in Robotics, AI, CS, Mechanical Engineering
PhD students and postdocs working on motion planning, control, or perception
Anyone passionate about humanoid robotics and climbing challenges

Practical details

Standard robotic platforms will be available on-site
Participants are welcome to bring their own robot
On-site accommodation available at the SMART facility

Contact & Registration

Interested in joining? Send us an email or subscribe to the mailing list for updates on registration, the simulator release, and the official rules.

Get in Touch

Venue

SMART Facility — University of Bordeaux

The workshop is held at the SMART (Sport, Motion, Arts & Research Technology) sports science research facility in Bordeaux, France. The site features a full motion capture studio, force platforms, and an indoor climbing wall — ideal for human climbing data collection and robot experimentation.

Complexe sportif SMART, Bordeaux, France