Research map / Reinforcement learning / Visual and representation RL
RRL: Resnet as representation for Reinforcement Learning
Rutav M Shah, Vikash Kumar
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 |
|---|---|---|---|
| Use frozen ImageNet-pretrained Resnet34 features in place of learned encoders (RRL+SAC) | SAC on DMControl · DMControl (6 tasks) | score | Hurt |
| Freeze RAD encoder pretrained on Cartpole and reuse across tasks | RAD encoder · DMControl (6 tasks) | score | Hurt |
| Use Resnet34 frozen features with sparse rewards and demonstrations (RRL) instead of dense shaped rewards (ablation) | RRL on Adroit · Adroit manipulation suite | task performance | No clear effect |
| Use ResNet34 instead of ResNet18 as frozen feature extractor (ablation) | RRL · Adroit manipulation suite | task performance | Improved |
| Use ResNet50 instead of ResNet34 as frozen feature extractor (ablation) | RRL · Adroit manipulation suite | task performance | Hurt |
| Use VAE trained on task-specific data instead of ImageNet ResNet features (ablation) | RRL · Adroit manipulation suite | task performance | Hurt |
| Use MobileNetV2 or ShuffleNet or VDVAE features instead of ResNet34 (ablation) | RRL · Adroit manipulation suite | task performance | No clear effect |
| Replace FERM with RRL (frozen pretrained ResNet features) on Adroit | RRL vs FERM · Adroit manipulation suite | task performance and compute | Improved |
| Add 2% noise to joint encoder signals or use vision-only inputs (ablation) | RRL · Adroit manipulation suite | task performance | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- CURL — CURL: Contrastive Unsupervised Representations for Reinforcement Learning (ICML 2020) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.88)
- ReinboT: Amplifying Robot Visual-Language Manipulation with Reinforcement Learning (ICML 2025 · similarity 0.87)
- Stabilizing Contrastive RL: Techniques for Robotic Goal Reaching from Offline Data (ICLR 2024 · similarity 0.87)
- Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers (NeurIPS 2024 · similarity 0.87)
- RLIF: Interactive Imitation Learning as Reinforcement Learning (ICLR 2024 · similarity 0.86)
- Generalizable Domain Adaptation for Sim-and-Real Policy Co-Training (NeurIPS 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.
