Skip to content
Changda Tian

Learning-Based Locomotion Control

Hierarchical RL Locomotion for a Six-Legged Robot

An early sim-to-real system: a reinforcement-learning planner on top of a trajectory-optimization gait controller for a hexapod.

Period
Sep 2019 – Apr 2020
Platform
Hexapod robot · Simulation
Methods
Reinforcement Learning · Trajectory Optimization · Sim-to-Real
With
Yue Gao (SJTU)
Paper
The six-legged robot used for the sim-to-real locomotion experiments

Overview

My first legged-robot system, built during my undergraduate years at SJTU: a two-level locomotion controller for a six-legged robot that separates where to go from how to step.

System

  • A top-level controller based on reinforcement learning performs path planning.
  • A low-level controller based on trajectory optimization plans gaits and foot positions.
  • Training happens entirely in simulation; the trained policy then drives the physical hexapod — an early, hands-on lesson in the sim-to-real gap.

Legacy

The layered structure of this project — learning on top, optimization below — became the template for my later work on capability-aware locomotion, where the interface between the layers is itself a learned model.

Publications

  • Sim-to-Real: Designing Locomotion Controller for Six-Legged Robot. Chenyu Yang, Yue Gao, Changda Tian, QingShan Yao. IEEE CYBER 2019, pp. 746–751. DOI