Researchers have introduced RCT, a new dataset collected by robots that focuses on tactile perception for generalization in object manipulation. The dataset comprises 29,279 tactile frames from 122 industrial reference materials, recorded using DIGIT sensors. RCT is designed to enable robust evaluation by preserving contact sequences, which is crucial for assessing how well robots can generalize to unseen materials and contact scenarios. The findings indicate that standard evaluation splits can be misleading due to strong correlations within presses, highlighting the challenge of novel-material generalization in robotic tactile perception. AI
IMPACT This dataset could improve robotic manipulation by enabling better generalization to new materials through tactile sensing.
RANK_REASON The cluster contains an academic paper describing a new dataset for research purposes.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
