Research map / Reinforcement learning / Visual and representation RL
TimeRewarder: Learning Dense Reward from Passive Videos via Frame-wise Temporal Distance
Yuyang Liu, Chuan Wen, Yihang Hu, Dinesh Jayaraman, Yang Gao
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 implicit negative sampling (reverse-ordered frame pairs) (ablation) | TimeRewarder · Meta-World stick-push and basketball | RL success rate | Improved |
| add weighted sampling favoring short intervals instead of uniform (ablation) | TimeRewarder · Meta-World stick-push and window-open | RL success rate | Improved |
| add two-hot discretization instead of direct regression (ablation) | TimeRewarder · Meta-World basketball and disassemble | RL success rate | Improved |
| vary number of discretization bins K (ablation) | TimeRewarder · Meta-World | RL success rate | No clear effect |
| vary reward combination factor alpha (ablation) | TimeRewarder · Meta-World | RL success rate | No clear effect |
| use frame-pair relative progress instead of only-from-init, single-frame input or order prediction (ablation) | TimeRewarder · Meta-World | RL success rate | Improved |
| add negative sampling to PROGRESSOR (ablation) | PROGRESSOR · Meta-World window-close and drawer-open | RL success rate | Mixed |
| use ResNet-34 instead of ViT-B backbone for VIP (ablation) | VIP · Meta-World | RL success rate | Mixed |
| combine human videos with Meta-World demonstration (ablation) | TimeRewarder · Meta-World cross-domain (3 tasks) | RL success rate | Improved |
| use from-scratch ViT-B instead of CLIP-pretrained ViT-B (ablation) | TimeRewarder · Meta-World | RL success rate | Hurt |
| use TimeRewarder instead of environment dense reward | DrQ-v2 · Meta-World (10 tasks) | RL success rate | Improved |
Closest papers by meaning
- Subtask-Aware Visual Reward Learning from Segmented Demonstrations (ICLR 2025 · similarity 0.89)
- SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Text2Reward: Reward Shaping with Language Models for Reinforcement Learning (ICLR 2024 · similarity 0.88)
- Learning Process Rewards via Visitation Matching for Efficient RL (NeurIPS 2026 · similarity 0.87)
- RoboAlign-R1: Distilled Multimodal Reward Alignment for Robot Video World Models (NeurIPS 2026 · similarity 0.87)
- Periodic Skill Discovery (NeurIPS 2025 · similarity 0.87)
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.
