Tacitus Robotics

We’re building dexterous robotic hands and the models that control them.

The goal is a hand that can apply what it learns to new tasks, rather than learning each task from scratch.

Current work

Our research combines predictive pretraining for control with a tendon-driven finger design. The mechanics and learning system are being developed together, with a focus on how the finger moves while staying in contact with an object.

CAD outline study Two-dimensional outlines extracted from the current single-finger CAD and full-hand reference assembly, with two actual finger tendon paths. The hand is viewed without its forearm housing. A fine line traces the geometry; this is not a physical simulation.