// the find
leggedrobotics/pace-sim2real
PACE: A systematic approach for sim-to-real transfer of legged robots, identifying actuator and joint dynamics with standard joint encoders.
PACE is a system identification framework for legged robots that uses CMA-ES to fit actuator and joint dynamics parameters from encoder data, then applies those parameters in NVIDIA Isaac Lab simulation to close the sim-to-real gap. It comes out of ETH Zurich's Robotic Systems Lab — the same group behind ANYmal — and targets researchers who already have a legged robot and an Isaac Lab setup. This is not a general robotics toolkit; it does one specific thing.
CMA-ES is a principled choice for this kind of black-box parameter optimization: gradient-free, handles non-convex landscapes, and the paper behind this has IJRR acceptance which means the method has been peer-reviewed rather than just shipped. The scope is narrow and honest — it models physically meaningful parameters (viscous friction, actuator dynamics) rather than learning a black-box residual, so you can inspect what it learned. Documentation is unusually complete for a research release: MkDocs site, tutorials, best-practices guide, and a worked ANYmal D example with concrete data collection scripts. Active ETH RSL maintenance with named maintainers who are reachable — this is not an abandoned paper dump.
Hard Isaac Sim 5.0 dependency is a real barrier: Isaac Sim is NVIDIA-only, license-encumbered, and requires a capable workstation GPU; anyone not already in that ecosystem cannot evaluate this at all. The framework is effectively single-robot validated in public form — ANYmal D is the only worked example, and transferring the config to a different robot requires understanding both Isaac Lab internals and PACE's parameter space, which the docs don't fully walk you through. CMA-ES convergence time is not discussed anywhere in the README or visible docs, and for high-dimensional actuator parameter spaces it can be slow; there's no guidance on what to expect or how to tune the optimizer budget. The 'active development, APIs may evolve' warning combined with an Isaac Lab version pin means you may find yourself debugging version drift rather than doing robotics research.