Research map / Label-efficient and robust learning / Recognition and long-tail
Embodied Representation Alignment with Mirror Neurons
Wentao Zhu, Zhining Zhang, Yuwei Ren, Yin Huang, Hao Xu, Yizhou Wang
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 mirror-neuron representation alignment (contrastive loss) to action recognition training | ViCLIP · RLBench action recognition (18 tasks) | accuracy: 71.6 -> 74.9 | Improved |
| add mirror-neuron alignment to embodied execution training | ARP · RLBench multi-task manipulation (18 tasks) | success rate: 85.3 -> 88.8 | Improved |
| add mirror-neuron alignment, fine-grained task Sort Shape | ARP · RLBench multi-task manipulation (18 tasks) - Sort Shape | success rate: 49.3 -> 66.7 | Improved |
| add mirror-neuron alignment, task Stack Block | ARP · RLBench multi-task manipulation (18 tasks) - Stack Block | success rate: 56.0 -> 72.0 | Improved |
| add mirror-neuron alignment, task Sort Shape | ViCLIP · RLBench action recognition (18 tasks) - Sort Shape | accuracy: 77.3 -> 85.3 | Improved |
| fine-tune ViCLIP on object manipulation data versus zero-shot | ViCLIP · RLBench action recognition (18 tasks) | accuracy: 10.4 -> 71.6 | Improved |
| use positive pairs by instruction instead of by class (tau 0.1) (ablation) | MN alignment · RLBench multi-task manipulation (18 tasks) (alignment ablation) | success rate: 85.7 -> 88.8 | Improved |
| use positive pairs by instruction instead of by class (tau 0.1) (ablation) | MN alignment · RLBench action recognition (18 tasks) (alignment ablation) | accuracy: 71.6 -> 74.9 | Improved |
| use positive pairs by instruction instead of by episode (tau 0.1) (ablation) | MN alignment · RLBench action recognition (18 tasks) (alignment ablation) | accuracy: 72.9 -> 74.9 | Improved |
| raise temperature from 0.1 to 0.2 with by-instruction positives (ablation) | MN alignment · RLBench action recognition (18 tasks) (alignment ablation) | accuracy: 74.9 -> 77.1 | Improved |
| raise temperature from 0.1 to 0.2 with by-instruction positives (ablation) | MN alignment · RLBench multi-task manipulation (18 tasks) (alignment ablation) | success rate: 88.8 -> 87.0 | Hurt |
| lower temperature from 0.1 to 0.02 with by-class positives (ablation) | MN alignment · RLBench multi-task manipulation (18 tasks) (alignment ablation) | success rate: 85.7 -> 86.9 | Improved |
Closest papers by meaning
- Bridging Critical Gaps in Convergent Learning: How Representational Alignment Evolves Across Layers, Training, and Distribution Shifts (NeurIPS 2025 · similarity 0.85)
- Mirror, Mirror on the Wall: Can VLM Agents Tell Who They Are at All? (NeurIPS 2026 · similarity 0.85)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · similarity 0.84)
- Towards a learning theory of representation alignment (ICLR 2025 · similarity 0.84)
- Neural Representational Consistency Emerges from Probabilistic Neural-Behavioral Representation Alignment (ICML 2025 · similarity 0.84)
- Selective Visual Representations Improve Convergence and Generalization for Embodied AI (ICLR 2024 · 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.
