Event
ForeTac: A Predictive Contact-Guided Framework Improves Robot Manipulation Success Rates
1 reports · 1 sources
Zhiyuan Chen and other researchers proposed a predictive contact-guided framework, ForeTac. On four real robot tasks, compared with the vision-only Diffusion Policy, the average task success rate increased from 36.3% to 71.3%, and the average contact success rate increased from 20.0% to 66.3%. ForeTac uses a representation-prediction-evaluation-guidance loop: it encodes the current tactile state, predicts future tactile latent trajectories according to candidate action sequences, and then evaluates each trajectory using a contact quality objective learned from successful and failed physical interactions. At inference time, the gradient of this objective is backpropagated through the tactile predictor to guide action generation before execution, while all model parameters remain frozen. The evaluation covers four tasks: flat and curved surface wiping, card swiping, and fragile object grasping, using diffusion policies and flow matching policies; under both policies, both metrics improved.
Report timeline
- ForeTac: A Predictive Contact Guidance Framework for Generative Robot Policies
ForeTac is a predictive contact guidance framework for contact-rich manipulation that uses a represent-predict-evaluate-guide loop: it encodes the current tactile state, predicts future tactile latent trajectories conditioned on candidate action sequences, and evaluates each trajectory with a contact-quality objective learned from successful and failed physical interactions. At inference, gradients of this objective are backpropagated through the tactile predictor to guide action generation before execution, while all model parameters remain frozen.
arXiv cs.RO ·