ForceBody dataset pairs parametric body motion with forces and torque uncertainty
Researchers released ForceBody, a dataset linking ground reaction forces and joint torques to the SKEL parametric body model, with per-sample uncertainty labels for torque estimates.
In plain words
A team built a dataset that attaches measured forces and computed joint torques to a standard digital body model used in motion research. Each torque label also comes with an uncertainty score, so users know how reliable it is. The data and code are publicly released.
What was announced
The authors say learning-based pipelines for body reconstruction and animation often lack force and torque data, and existing motion datasets with force annotations do not fit parametric body models. ForceBody addresses this by pairing the SKEL body model with ground reaction forces and inverse-dynamics joint torques drawn from 10,386 motion trials, covering 9.7 million frames and 26.9 hours from 140 subjects. Torque labels are derived through inverse dynamics, a method the authors describe as sensitive to the quality of input motion data, so each label carries a per-frame, per-joint uncertainty annotation computed via Monte Carlo sampling through the pipeline.
To demonstrate use, the team benchmarked six architectures — MLP, Conv1D, LSTM, GRU, Mamba2 and Transformer — on predicting ground reaction forces and joint torques from SKEL motion. In one example, finetuning a Transformer with per-sample uncertainty weighting reduced the mean absolute error of joint torque estimation by 5%. Code and data are available in a public repository.
What it does not show
- The evidence is the paper abstract only; the full paper's methods, validation and peer review status are not covered here.
- The 5% error reduction is the authors' own reported benchmark result and has not been independently confirmed.
For security and IT teams
- Access
- The dataset and code are publicly available at the repository linked in the abstract.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.