Research map / Reinforcement learning / Visual and representation RL
From Virtual Demonstration to Real-World Manipulation Using LSTM and MDN
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 LSTM-MDN instead of Feedforward-MSE (ablation) | robot controller · Pick and place (virtual world) | success rate: 0% -> 100% | Helped |
| Use LSTM-MDN instead of Feedforward-MSE (ablation) | robot controller · Push to pose (virtual world) | success rate: 0% -> 95% | Helped |
| Add LSTM layers to an MDN controller (ablation) | robot controller · Pick and place (virtual world) | success rate: 95% -> 100% | Helped |
| Add LSTM layers to an MDN controller (ablation) | robot controller · Push to pose (virtual world) | success rate: 15% -> 95% | Helped |
| Use an MDN output and likelihood loss instead of MSE (ablation) | LSTM controller · Pick and place (virtual world) | success rate: 85% -> 100% | Helped |
| Use an MDN output and likelihood loss instead of MSE (ablation) | LSTM controller · Push to pose (virtual world) | success rate: 0% -> 95% | Helped |
| Transfer the unchanged virtual-trained LSTM-MDN controller to the physical robot | LSTM-MDN · Pick and place (physical world) | success rate: 100% -> 80% | Hurt |
| Transfer the unchanged virtual-trained LSTM-MDN controller to the physical robot | LSTM-MDN · Push to pose (physical world) | success rate: 95% -> 60% | Hurt |
| Include imperfect demonstrations containing errors and human recovery actions | LSTM-MDN controller · Manipulation tasks in virtual and physical environments | self-correction | Helped |
Closest papers by meaning
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning (ICML 2025 · similarity 0.84)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · similarity 0.84)
- Generalizable Domain Adaptation for Sim-and-Real Policy Co-Training (NeurIPS 2025 · similarity 0.84)
- HAMSTER: Hierarchical Action Models for Open-World Robot Manipulation (ICLR 2025 · similarity 0.83)
- ManiSkill-HAB: A Benchmark for Low-Level Manipulation in Home Rearrangement Tasks (ICLR 2025 · similarity 0.83)
- DexTrack: Towards Generalizable Neural Tracking Control for Dexterous Manipulation from Human References (ICLR 2025 · similarity 0.83)
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.