Research map / Probabilistic and scientific modeling / Bayesian optimization
Using Simulation to Improve Sample-Efficiency of Bayesian Optimization for Bipedal Robots
Akshara Rai, Rika Antonova, Franziska Meier, Christopher G. Atkeson
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add DoG-based simulation kernel to BO | BO with SE kernel · ATRIAS hardware 5D controller | runs finding walking: SE: 60% in 10 trials, 80% in 20 trials; DoG: 100% within 3 trials | Improved |
| add DoG-based simulation kernel to BO | BO with SE kernel · ATRIAS hardware 9D controller | runs finding walking: SE: 0/3 in 10 trials; DoG: 3/3 in 5 trials | Improved |
| add NN-based trajectory-summary kernel to BO | BO with SE kernel · ATRIAS hardware 9D controller (IMU restored) | runs finding walking: SE: 2 of 5 runs in 10 trials; NN: 5 of 5 within 6 trials | Improved |
| add DoG or NN kernel to BO under smooth cost | BO with SE kernel · 7-link biped 16D NMC, smooth cost | runs achieving walking: 90% runs after 25 trials vs 90 trials for SE | Improved |
| add DoG or NN kernel to BO under non-smooth cost | BO with SE kernel · 7-link biped 16D NMC, non-smooth cost | runs finding walking after 100 trials: SE: 1 out of 50 runs; NN: 70% | Improved |
| add hardware mismatch adjustment to DoG kernel (ablation) | DoG kernel · ATRIAS simulation 50D VNMC (original simulator) | walking found / performance: both find walking in 20 trials; adjusted performs better | Improved |
| add mismatch adjustment to DoG kernel at larger simulator mismatch (no boom) (ablation) | DoG kernel · ATRIAS simulation 50D VNMC (no boom, simplified gear simulator) | runs finding walking: approx. 90% after 50 trials -> all runs after 35 trials | Improved |
| use informed kernels at low mismatch (simplified gear dynamics) | BO with SE kernel · ATRIAS simulation 50D VNMC (simplified gear dynamics simulator) | runs with walking after 50 trials: 70% -> all runs | Improved |
| add simulated cost prior built from a simulator with no boom | BO with SE kernel · ATRIAS simulation 50D VNMC | walking performance | Hurt |
| add simulated cost prior built from simplified gear dynamics simulator | BO with SE kernel · ATRIAS simulation 50D VNMC | walking performance | Improved |
| use IT&E with no-boom simulator map | BO with SE kernel · ATRIAS simulation 50D VNMC | runs finding walking in 50 trials: 77% of runs in 50 trials | Mixed |
| add cost prior to IT&E (versus kernel-only IT&E) (ablation) | IT&E · ATRIAS simulation 50D VNMC | walking performance | Improved |
Closest papers by meaning
- Augmented Bayesian Policy Search (ICLR 2024 · similarity 0.86)
- Model-based Diffusion for Trajectory Optimization (NeurIPS 2024 · similarity 0.86)
- Scalable Bayesian Optimization via Focalized Sparse Gaussian Processes (NeurIPS 2024 · similarity 0.85)
- Efficient Test-Time Adaptation For Robot Policies (NeurIPS 2026 · similarity 0.84)
- ASID: Active Exploration for System Identification in Robotic Manipulation (ICLR 2024 · similarity 0.84)
- Rapidly Adapting Policies to the Real-World via Simulation-Guided Fine-Tuning (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
