Sanctuary has demonstrated sim-to-real transfer of learned manipulation policies for its hydraulic hands. | Source: Sanctuary AI
Sanctuary AI this week shared that it uses NVIDIA Isaac Lab to simulate dexterity-focused training environments. Isaac Lab is an open-source, unified framework that enables training of robot policies with high-fidelity simulation.
Built on NVIDIA Isaac Sim, Isaac Lab uses PhysX for physics simulation and RTX rendering to bridge the gap between simulation and perception-based robot training. This can help researchers and developers build autonomous robots more efficiently, according to Sanctuary AI.
The Vancouver, Canada-based company uses Isaac Lab to train its hydraulic hands to perform cutting-edge manipulation tasks. It does this first in simulation, and ultimately in reality. Sanctuary AI asserted that its hands have kinematics beyond human capability, which cannot be accessed using analogous teleoperation.
Online reinforcement learning in a simulated environment allows the learning algorithms to fully use the hands’ capabilities. In addition, Sanctuary AI said it can use simulation to train thousands of hands at once, drastically accelerating the learning process.
Founded in 2018, Sanctuary Cognitive Systems Corp. has been recognized as a leader in intellectual property around general-purpose robots and embodied artificial intelligence. Morgan Stanley recently ranked it third globally for published U.S. patents.
Sanctuary unveiled the latest iteration of its Phoenix robots last year and raised funding, bringing its total to $140 million. The company also demonstrated in-hand object manipulation with its 21-degree-of-freedom (DoF) hand.
Sanctuary AI adds tactile sensing, hydraulic actuation
Sanctuary’s proprietary robotic gripper is differentiated by a high number of active degrees of freedom. This allows for finger abduction and therefore in-hand manipulation, a feature it said other commercially available hands do not offer.
In February, the company integrated new tactile sensors into its Phoenix robots. Sanctuary said this will enable teleoperated pilots to more effectively conduct complex, touch-driven tasks with precision and accuracy. It began working on tactile sensing with its acquisition of Tangible Research two years ago.
Sanctuary claimed that Phoenix is unique in its use of hydraulic actuation, providing high strength, speed, and controllability. The company said it believes miniaturized hydraulic valves provide a viable path to achieving human-level dexterity.
By contrast, Boston Dynamics used hydraulic actuation in its previous Atlas models but switched to electric actuation when it decided to create a commercial model.
More humanoid news from GTC
Sanctuary AI’s news was part of several humanoid-related announcements from NVIDIA GTC this week. First, Agility Robotics announced companies also are collaborating to make Digit models available to partners through “Mega,” an NVIDIA Omniverse Blueprint.
In a separate announcement, Boston Dynamics has expanded its collaboration with NVIDIA to build the next generation of AI capabilities for humanoid robots. As an early adopter of the NVIDIA Isaac GR00T platform, Boston Dynamics’ Atlas robot is one of many humanoids using the NVIDIA Jetson Thor computing platform.
Tell Us What You Think!