Differentiable Co-Optimization of Mass-Limited Legged Robots
Project Overview
This project investigates how morphology and locomotion control can be jointly optimized for legged robots operating across diverse gravitational environments, treating the robot's body as an active participant in task performance rather than a fixed platform for control. By propagating gradients through long-horizon rigid-body contact simulations, we study how body geometry, mass allocation, and gait coordination reorganize together as gravity and actuator authority change. This represents a form of embodied intelligence where adaptive behavior emerges from the coupling between body design, actuation, and the physical environment.
Building on earlier work in the ESA Advanced Concepts Team on Space Gaits, which established gravity and terrain as regime-defining variables for legged gait analysis, this project moves from gait selection on fixed bodies to full morphology-gait co-optimization using differentiable physics simulation.
Key findings include gravity-dependent mass redistribution between body and limbs, the spontaneous emergence of vestigial legs as an optimizer strategy under high gravity and limited actuation, and systematic gait regime transitions from aerial to ground-bound coordination - all arising without manual design intervention.
This work supports ESA's Strategy 2040 objectives for autonomy and innovation in space robotics, and contributes to the foundations of the upcoming Embodied Intelligence for Space Robotics initiative, which aims to develop robotic systems capable of autonomous reasoning and dynamic adaptation in unstructured extraterrestrial environments.
Further Reading
- NVIDIA, "Warp: A Python framework for high performance GPU simulation and graphics," 2022.
- E. Papadopoulos et al., "Ariadna Study: Space Gaits (Final Report)," ESA Advanced Concepts Team, 2013.
- P. Arm et al., "Scientific exploration of challenging planetary analog environments with a team of legged robots," Science Robotics, 2023.