AuSRoS Hands-on Workshop: High-level Planning to Low-level Control
Hands-on workshop material connecting RRT/MCTS planning with learned locomotion policies in Isaac Sim and Isaac Lab.
Work in progress. This page is currently a lightweight overview while the workshop code and participant instructions continue to evolve in the repository.
This project contains teaching material and demonstration code for an AuSRoS hands-on workshop on connecting high-level planning to low-level robot control. The workshop uses Isaac Sim and Isaac Lab to show how planning algorithms such as RRT and MCTS can be connected to learned locomotion policies for simulated legged robots.
The main teaching message is that a valid geometric path is not necessarily executable robot behaviour. Execution also depends on waypoint tracking, command generation, obstacle clearance, controller limits, dynamics, and the behaviour of the learned policy.
Links
Key Topics
Robotics workshop · Isaac Sim · Isaac Lab · Learned locomotion · RRT · MCTS · Kinodynamic planning · Simulation
Workshop Goal
Connect planning algorithms to robots running in simulation by bridging the gap between high-level geometric path planning and low-level dynamic execution.
What Participants Explored
- Planning with learned locomotion in Isaac Sim, including RRT paths executed by bipedal and quadrupedal robots.
- Training and understanding locomotion policies in Isaac Lab, and how those policies shape what a robot can actually execute.
- Sparse versus dense waypoint following, path-to-command tracking, and why a geometric path still needs a control interface.
- Kinodynamic planning, kinematic rollouts versus policy-in-the-loop rollouts, and offline planning versus online MCTS replanning.
Key Takeaway
A geometric path is not yet robot behaviour. Reliable execution emerges only when planning, waypoint tracking, learned locomotion policies, and robot dynamics are designed to work together.