Research map / Reinforcement learning / Visual and representation RL
Memory, Benchmark & Robots: A Benchmark for Solving Complex Tasks with Reinforcement Learning
Egor Cherepanov, Nikita Kachaev, Alexey Kovalev, Aleksandr Panov
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 |
|---|---|---|---|
| increase action chunk size from K=4 to K=8 | OpenVLA · MIKASA-Robo ShellGameTouch | success rate: 0.12 -> 0.47 | Improved |
| increase action chunk size from K=4 to K=8 | OpenVLA · MIKASA-Robo RememberColor3 | success rate: 0.21 -> 0.59 | Improved |
| increase action chunk size from K=4 to K=8 | OpenVLA · MIKASA-Robo RememberColor9 | success rate: 0.08 -> 0.06 | No clear effect |
| increase number of colors from 3 to 9 | Octo-small · MIKASA-Robo RememberColor | success rate: 0.45 -> 0.11 | Hurt |
| increase number of colors from 3 to 9 | pi0 (K=4) · MIKASA-Robo RememberColor | success rate: 0.35 -> 0.15 | Hurt |
| add LSTM memory to PPO | PPO-MLP · MIKASA-Robo RememberColor3 (RGB+joints, dense reward) | success rate: 25% -> 100% | Improved |
| increase number of cubes from three to five or nine | PPO-LSTM · MIKASA-Robo RememberColor (dense reward) | success rate | Hurt |
| switch from dense to sparse reward | PPO-MLP and PPO-LSTM · MIKASA-Robo RememberColor3 | success rate | Hurt |
| provide full state mode instead of RGB+joints | PPO-MLP · MIKASA-Robo all tasks | success rate: 25% -> 100% | Improved |
| use RATE (memory transformer) instead of DT | RATE vs DT · MIKASA-Robo RememberColor3 (offline RL, dense reward) | success rate: 0.05 -> 0.68 | Improved |
| use RATE instead of CQL | RATE vs CQL · MIKASA-Robo ShellGameTouch (offline RL, dense reward) | success rate: 0.02 -> 0.97 | Improved |
| use RATE instead of BC | RATE vs BC · MIKASA-Robo RememberColor9 (offline RL, dense reward) | success rate: 0.15 -> 0.10 | No clear effect |
Closest papers by meaning
- SAM2Act: Integrating Visual Foundation Model with A Memory Architecture for Robotic Manipulation (ICML 2025 · similarity 0.88)
- Stable Hadamard Memory: Revitalizing Memory-Augmented Agents for Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Benchmarking Agent Memory in Interdependent Multi-Session Agentic Tasks (ICML 2026 · similarity 0.88)
- VLA-Pro: Cross-Task Procedural Memory Transfer for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.87)
- RoboCerebra: A Large-scale Benchmark for Long-horizon Robotic Manipulation Evaluation (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.
