Research map / Reinforcement learning / Visual and representation RL
Unsupervised Domain Adaptation with Dynamics-Aware Rewards in Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train skills in simulation with dynamics-aware rewards and limited real-world interaction | Quadruped sim2real · Moving forward and backward | time to valid skill emergence (hours): >6 -> 3 | Helped |
| Train skills in simulation with dynamics-aware rewards and limited real-world interaction | Quadruped sim2real · Keeping balance | time to valid skill emergence (hours): >6 -> 1; 4 -> 1 | Helped |
| Add target-dynamics KL reward modification to source-trained skills | DARS · Stable map transfer tasks | goal-reaching performance | Helped |
| Reuse target-environment data in policy updates (ablation) | DARS · Stable map transfer tasks | goal-reaching performance | No clear effect |
| Increase the target-environment GPIM training budget tenfold instead of using DARS | GPIM · Stable map transfer tasks | goal-reaching performance | Hurt |
| Finetune source-trained GPIM in the target instead of using DARS | GPIM · Stable map transfer tasks | goal-reaching performance | Hurt |
| Use DARS instead of GPIM trained in the target | Goal-conditioned policy · Stable map transfer tasks | goal-reaching performance | Mixed |
| Use DARS instead of SMiRL or SMiRL finetuning for dynamics-shifted unstable environments | Unstable humanoid transfer (H, A-H) and (H, B-H) · Humanoid stability | skill stability | Helped |
| Increase the KL regularization coefficient; an overly high setting uses β = 50 (ablation) | DARS · Dynamics-shift transfer tasks | target performance: β = 50 | Mixed |
Closest papers by meaning
- Off-Dynamics Reinforcement Learning via Domain Adaptation and Reward Augmented Imitation (NeurIPS 2024 · similarity 0.90)
- Cross-Domain Policy Adaptation by Capturing Representation Mismatch (ICML 2024 · similarity 0.90)
- In-Context Policy Adaptation via Cross-Domain Skill Diffusion (AAAI 2025 · similarity 0.89)
- Constrained Ensemble Exploration for Unsupervised Skill Discovery (ICML 2024 · similarity 0.89)
- Robust Policy Learning via Offline Skill Diffusion (AAAI 2024 · similarity 0.89)
- Efficient Skill Discovery via Regret-Aware Optimization (ICML 2025 · similarity 0.89)
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.