Research map / Vision-language and multimodal / Robot action and navigation
EgoRoC: Towards Egocentric Robotic Control via Task-Agnostic Visual Alignment
Wei Feng, Chi Zhang, Nan Li, Qian Zhang, Qi Zhang, Mingyan Li
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 |
|---|---|---|---|
| attach EgoRoC alignment head to five VLA baselines | OpenVLA, TinyVLA, SpatialVLA, Pi-0, OpenVLA-OFT · Isaac Sim and real world, OOD tasks | success rate (%) | Improved |
| attach EgoRoC during fine-tuning with same demonstrations | OpenVLA and Pi-0 · real-world Task-L, Po, PA, PD | success rate (%) | Improved |
| train DHCM with second stage in addition to first stage (ablation) | DHCM · real-world Task-L | success rate (%): 66.0 -> 72.0 | Improved |
| replace Tsai-Lenz with fully trained DHCM (ablation) | EgoRoC · real-world Task-PD | success rate (%): 17.0 -> 20.0 | Improved |
| use first-stage-only DHCM instead of Tsai-Lenz (ablation) | EgoRoC · real-world Task-L | success rate (%): 67.0 -> 66.0 | Hurt |
| add grid-based COTR matching (4x4) (ablation) | DHCM · real-world Task-L | success rate (%): 61.0 -> 72.0 | Improved |
| increase grid from 4x4 to 8x8 (ablation) | DHCM · real-world Task-L | success rate (%): 72.0 -> 77.0; speed 0.19 s -> 0.87 s | Mixed |
| add Rank-N-Contrast loss (ablation) | DHCM · hand-eye calibration | rotation angle error (deg): 12.78 -> 2.03 | Improved |
| replace MLP regression with diffusion generator (ablation) | DHCM · hand-eye calibration | rotation angle error (deg): 6.27 -> 2.03 | Improved |
| add Geo-MLP (ablation) | DHCM · real-world Task-PD | success rate (%): 17.0 -> 20.0 | Improved |
| add Geo-MLP (ablation) | DHCM · real-world Task-Po | success rate (%): 52.0 -> 52.0 | No clear effect |
| share OpenVLA LMM for EVAM instead of separate Qwen2.5-VL-7B (ablation) | EgoRoC with OpenVLA · real-world average | success rate (%): 15.5 -> 15.8 | No clear effect |
Closest papers by meaning
- Pose6DAug: Physically Plausible Multi-View Object Swapping for Robot Data Augmentation (NeurIPS 2026 · similarity 0.88)
- OASIS: Observation-Action Space Alignment via SE(3) Trajectory Prediction for Robotic Manipulation (NeurIPS 2026 · similarity 0.88)
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.87)
- HandEdit: A Unified Benchmark for Egocentric Human-to-Robot Dexterous Hand Image Editing (NeurIPS 2026 · similarity 0.87)
- Robotic Visual Instruction (CVPR 2025 · similarity 0.87)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · 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.
