Research map / Vision-language and multimodal / Robot action and navigation
RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies
Yinpei Dai, Hongze Fu, Jayjun Lee, Yuejiang Liu, Haoran Zhang, Jianing Yang, Chelsea Finn, Nima Fazeli, Joyce Chai
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 perceptual memory (frame sampling, memory-as-modulator) to pi0.5 | pi0.5 · RoboMME (16 tasks, average success) | success rate (%): 17.93 -> 44.51 | Improved |
| use frame sampling instead of token dropping for perceptual memory (Modul) (ablation) | perceptual memory · RoboMME (16 tasks, average success) | success rate (%): 38.04 -> 44.51 | Improved |
| integrate frame-sampled memory as modulator instead of context (ablation) | FrameSamp · RoboMME (16 tasks, average success) | success rate (%): 30.68 -> 44.51 | Improved |
| integrate frame-sampled memory as modulator instead of expert (ablation) | FrameSamp · RoboMME (16 tasks, average success) | success rate (%): 36.25 -> 44.51 | Improved |
| use recurrent memory (TTT or RMT) instead of perceptual memory (ablation) | pi0.5 memory · RoboMME (16 tasks, average success) | success rate (%): 44.51 -> 22.35 (best recurrent TTT+Expert) | Hurt |
| add RMT recurrent memory to pi0.5 | pi0.5 · RoboMME (16 tasks, average success) | success rate (%): 17.93 -> 20.17 (RMT+Modul) | No clear effect |
| use grounded subgoals instead of simple subgoals (QwenVL) (ablation) | symbolic memory · RoboMME (16 tasks, average success) | success rate (%): 29.00 -> 32.70 | Improved |
| use grounded instead of simple subgoals on simple tasks such as PickXTimes (ablation) | symbolic memory · RoboMME PickXTimes (QwenVL) | success rate (%): 95.33 -> 92.67 | Hurt |
| fine-tune Qwen3-VL-4B subgoal predictor instead of prompting Gemini-2.5-Pro (ablation) | GroundSG · RoboMME (16 tasks, average success) | success rate (%): 23.25 -> 32.70 | Improved |
| use oracle subgoals instead of VLM-predicted subgoals (ablation) | GroundSG · RoboMME (16 tasks, average success) | success rate (%): 32.70 -> 84.08 | Improved |
| append past actions to language tokens as symbolic memory | pi0.5 · RoboMME (16 tasks, average success) | success rate (%): 17.93 -> 19.73 | No clear effect |
| add historical proprioceptive states to perceptual memory (ablation) | perceptual memory · RoboMME | success rate (%) | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SAM2Act+ — SAM2Act: Integrating Visual Foundation Model with A Memory Architecture for Robotic Manipulation (ICML 2025) · better on 1 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
- SAM2Act: Integrating Visual Foundation Model with A Memory Architecture for Robotic Manipulation (ICML 2025 · similarity 0.92)
- RoboProcessBench: Benchmarking Process-Aware Understanding in Visual-Language Robotic Manipulation (NeurIPS 2026 · similarity 0.90)
- Exploring the Limits of Vision-Language-Action Manipulation in Cross-task Generalization (NeurIPS 2025 · similarity 0.90)
- RoboCerebra: A Large-scale Benchmark for Long-horizon Robotic Manipulation Evaluation (NeurIPS 2025 · similarity 0.90)
- VLA-Pro: Cross-Task Procedural Memory Transfer for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
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.
