Technology

Robotiq Releases New 2F-85 Gripper Isaac Sim Assets on Newton

Robotiq released updated Isaac Sim assets for the 2F-85 gripper, covering PhysX and Newton physics backends, addressing closed-loop kinematics simulation issues with documentation.

Published · Updated · Source: Robotiq · 协作机器人 · AI translation
Robotiq has released updated Isaac Sim assets for the 2F-85 Adaptive Gripper, supporting both PhysX and Newton physics backends, with documentation explaining the mechanism. Previously, when a 2F-85 was placed into an Isaac Sim scene and closed on an object, the fingers often exhibited behaviors the real gripper never does: bending backwards, pivoting outward at the tip, or snapping out of alignment mid-grasp, which is the most reported problem from the community. The gripper is underactuated: each finger is a five-bar linkage in which two links form a virtual extendable link, with a pin limiting the minimum length; at minimum length the fingertip stays parallel, and when the virtual link extends, the fingertip rotates inward to produce an encompassing grip. A torsion spring at the inner finger joint provides a passive degree of freedom, allowing the gripper to conform to round objects. Isaac Sim solves an articulation as a tree of rigid bodies, which cannot contain loops, but the 2F-85 finger is a loop. The public asset offers two USD variants: cutting the loop and re-imposing it algebraically with mimic constraints, which can only reproduce the parallel grip; or keeping every joint and closing the loop as a constraint, marking one joint as excludeFromArticulation and letting the solver enforce it in maximal coordinates, preserving the underactuated degree of freedom. Under PhysX, this closing constraint is a best-effort constraint, so under certain loads it tolerates misalignment, accumulates error, and the linkage snaps out of alignment. Newton, built on MuJoCo-Warp, models a closed loop natively as an equality constraint, keeping the five-bar closure stable at joint limits, solving all 10 joints analytically, and allowing the parallel-grip spring to hold so the finger returns to parallel when released and rotates inward when demanded, enabling encompassing grip simulation in the assets for the first time. Users wanting a fast, robust parallel pinch can choose the PhysX variant, while those simulating grip realism can choose Newton. Next are the 2F-140, then Hand-E in all its variants, with work toward validation against the SimReady Foundation specification. These assets are one of three open-source developer pieces, alongside the C++ SDK and ROS 2 packages. Geometry and images are in Git LFS, and the .usda layers are plain text. After installation, refer to the gripper simulation guide and the Newton asset's README. To mount on a robot, reference the config file for the chosen physics variant, attach it to the wrist link with a fixed joint, and remove the gripper's own articulation root so the arm and gripper solve as one articulation. To switch fingertips at integration time, reference Robotiq_2F_85_edit.usd. Under the compliant and Newton variants, command only finger_joint and let the loop closure solve the rest. Issues and feedback are welcome on GitHub. Information comes from the original source and is organized by AI.
Read the original source

Related news