DeCAL
AureleoPKU
A dexterous vision-language-action model combining contact-aware multimodal prediction and action generation.
What you can explore
Explore data conversion, training launchers, offline evaluation and server-based inference.
Before you use it
The pipeline needs multiple pretrained components and CUDA-specific dependencies. Training and inference code do not establish drop-in compatibility with another tactile hand.
Original contributors
Yankai Fu, Ning Chen and DeCAL collaborators
License / access: Code: Apache 2.0 · check external data, models and hardware terms separately
Editorial reference · checked 2026-09-22