Meta open sources Project SuperDex for robot dexterity
In a LinkedIn post, robotics commentator Lukas M. Ziegler said Meta Reality Labs Research has open sourced Project SuperDex, a robot dexterity stack built around contact focused physics simulation. According to Ziegler, the project is available on GitHub. The post is independent commentary rather than a Meta announcement, and the supplied material does not independently verify its technical claims.
Project SuperDex reportedly uses one solver for rigid bodies, soft bodies, rods, tendons, shells and cloth. Ziegler said the engine supports collision handling for nonconvex objects with detailed contact force distributions, along with tactile sensors and soft contact as native primitives. Constraint aware inverse kinematics runs on the same optimization core as forward dynamics, while the solver is claimed to remain stable without the tight time step limits associated with explicit methods.
The accompanying data workflow uses a Quest 3 headset to teleoperate a simulated hand with haptic feedback, allowing demonstrations to be collected without physical robot hardware. Ziegler also described a shape sorting policy trained entirely in simulation and deployed zero shot on a real robotic hand. No benchmark results or independent validation were included in the post. For manipulation teams, the practical proposition behind Project SuperDex is clear: collect contact rich demonstrations in simulation, then test whether they transfer to hardware.
Source: Lukas M. Ziegler on LinkedIn
