Reproduce “RoboTidy : A 3D Gaussian Splatting Household Tidying Benchmark for Embodied Navigation and Action” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiaoquan Sun, Ruijian Zhang, Kang Pang, Bingchen Miao, Yuxiang Tan, Zhen Yang, Ming Li, Jiayu Chen
RoboTidy introduces a benchmark for language-guided household tidying, utilizing photorealistic 3D Gaussian Splatting scenes and a physically accurate simulator to evaluate embodied agents in navigation and manipulation. It demonstrates robust sim2real transfer, improving real-world manipulation success rates by approximately 20% by combining synthetic and real demonstrations.
Key insights
- Advanced Vision-Language Models (VLMs) can effectively abstract object sorting rules from multimodal observations, outperforming simpler baselines in personalized tidying tasks.
- The diverse and physics-aware trajectories provided by the RoboTidy benchmark enhance VLA model robustness and generalization across unseen scenes and objects.
- Photorealistic 3DGS scenes and physically consistent simulation are crucial for narrowing the sim-to-real gap, enabling learned policies to transfer effectively to real-world robots.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.