Research map / Vision-language and multimodal / Cross-modal learning
Cross-modal Domain Adaptation for Cost-Efficient Visual Reinforcement Learning
Xiong-Hui Chen, Shengyi Jiang, Feng Xu, Zongzhang Zhang, Yang Yu
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 |
|---|---|---|---|
| replace MLP mapping function with RNN mapping function (ablation) | CODAS · Hopper and Inverted Double Pendulum | reward ratio | Improved |
| feed trajectories into the discriminator (RNN-N to RNN-Y) (ablation) | CODAS · Inverted Double Pendulum and Walker2d | reward ratio | Improved |
| add embedded-DM to RNN mapping (RNN-Y to DM-RNN-Y) (ablation) | CODAS · Hopper, Inverted Double Pendulum, Walker2d | reward ratio | Improved |
| add cycle-consistent loss to GAN (CycleGAN vs GAN) (ablation) | GAN baseline · MuJoCo tasks | reward ratio | No clear effect |
| stack input frames in GAN (GAN STACK vs GAN) | GAN baseline · MuJoCo tasks | reward ratio | Mixed |
| increase friction in source domain to 2.0x (dynamics mismatch) (ablation) | CODAS · Hopper | reward ratio | Hurt |
| increase friction in source domain to 1.2x and 1.6x (ablation) | CODAS · Hopper | reward ratio | No clear effect |
| degrade data-collecting policy to 77% and 46% of original performance (policy mismatch) (ablation) | CODAS · Hopper | reward ratio | Hurt |
| goal-conditional relocate task instead of other hand tasks | CODAS · Hand manipulation | reward ratio: 0.886 -> 0.090 | Hurt |
| use CODAS rather than GAN or CycleGAN with cross-modal mapping | CODAS · MuJoCo tasks | reward ratio | Improved |
| use CODAS rather than baselines for state mapping | CODAS · MuJoCo tasks | RMSE of mapped states | Improved |
| behavior cloning versus domain adaptation in HalfCheetah | BC · HalfCheetah | reward ratio | Mixed |
Closest papers by meaning
- Prompt-based Visual Alignment for Zero-shot Policy Transfer (ICML 2024 · similarity 0.88)
- Cross-Domain Policy Adaptation by Capturing Representation Mismatch (ICML 2024 · similarity 0.87)
- Generalizable Domain Adaptation for Sim-and-Real Policy Co-Training (NeurIPS 2025 · similarity 0.87)
- In-Context Policy Adaptation via Cross-Domain Skill Diffusion (AAAI 2025 · similarity 0.87)
- When Should We Prefer State-to-Visual DAgger over Visual Reinforcement Learning? (AAAI 2025 · similarity 0.86)
- Fourier Guided Adaptive Adversarial Augmentation for Generalization in Visual Reinforcement Learning (AAAI 2025 · similarity 0.86)
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.
