EgoTac model predicts tactile data from egocentric videos

Researchers have developed EgoTac, a novel model capable of predicting tactile information directly from egocentric human videos. Trained on a dataset of over 5.7 million image-tactile pairs, EgoTac demonstrates the ability to capture nuanced touch dynamics and continuous force measurements. The model achieves an average force error below 0.06N in domain-specific predictions and surpasses existing state-of-the-art methods for out-of-domain contact prediction. This advancement offers a scalable method for extracting tactile priors from readily available human video data, paving the way for more tactile-aware robot learning applications. AI

IMPACT Enables more tactile-aware robot learning by extracting priors from readily available egocentric human videos.

RANK_REASON The item is a research paper detailing a new model and its performance on specific benchmarks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• Connected Papers
• CORE Recommender
• DagsHub
• EgoTac
• Gotit.pub
• Hugging Face
• Litmaps
• ScienceCast
• scite Smart Citations

• paper
• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →