Researchers have introduced ForceBody, a novel dataset designed to integrate ground truth force and torque data into parametric body models for motion generation and reconstruction. This dataset addresses the limitations of existing motion pipelines by providing paired motion sequences with force annotations and quantifying the reliability of estimated joint torques. ForceBody includes over 10,000 motion trials from 140 subjects, featuring per-sample torque uncertainty annotations derived from Monte Carlo sampling. The dataset has been used to benchmark six different neural network architectures, demonstrating that incorporating per-sample uncertainty can improve joint torque estimation accuracy. AI
IMPACT This dataset could advance physics-informed AI for realistic human motion generation and robotics applications.
RANK_REASON The cluster describes a new dataset and benchmarks for a specific research area in computer vision and robotics, published on arXiv.
Read on Hugging Face Daily Papers →
- Conv1d
- DominoAI-Lab
- ForceBody
- gated recurrent unit
- long short-term memory
- Mamba2
- multilayer perceptron
- transformer
- arXiv
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →