Robotiq Inc.

Tactile sensing in robotics is not new. What is new is a gripper that interprets contact itself, on the tool, at the moment it happens. The industry is instrumenting vision, demonstrations and simulation. All three miss the moment that decides whether a grasp holds. That information sits at the contact patch, sampled far faster than any vision loop runs, and behind the fingers that are occluding it. The 2FC from @Robotiq_Inc puts the interpretation where the contact is: → on-gripper skills: grasp confidence, incipient slip detection, in-hand object pose, object centering, grip verification → 28 taxels per fingertip (4×7), 1000 Hz dynamic sensing, 3-axis IMU → pressure, vibration, proprioception, force, speed and position, timestamped to the millisecond in one reference frame Look again at that first line. Incipient slip detection and grip verification are not only runtime safety, they are automatic labelers. Every production pick writes a labeled episode with nobody annotating it. You cannot do millisecond labeling over a network, which is why the compute sits on the tool. In-hand pose estimation is the other tell. Underactuated fingers adapt mechanically, which is what you want for contact-rich work, but it means finger and object pose stop being inferable from the actuator alone. The tactile layer closes the observability gap the mechanics opened. Then train and deploy on the same hardware. 2M cycles, and the field upgrade preserves the TCP, so 30,000 deployed 2F grippers can become a data fleet without re-teaching a single program. On resolution: 28 taxels is coarse next to optical tactile arrays. That is a deliberate trade for 1 kHz bandwidth and a 2M-cycle duty rating, and there is published work doing texture recognition on these sensors regardless. Touch has been the modality nobody productised. This is the most serious attempt at it I have seen. :) 🔗 Learn more here: robotiq.com/physical-ai-grip… ~~ ♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com

If you've ever run a Robotiq gripper in ROS 2, someone wrote that driver on their own time 🦾 @Robotiq_Inc grippers have worked in ROS 2 and Isaac Sim for years, and the community put them there. Volunteers wrote the drivers, built the simulation models and kept them maintained, so the hardware would run inside the tools they were already using. This week Robotiq took ownership. Three open-source releases, all maintained by the company, all on github.com/robotiq: → An official ROS 2 package, built as a drop-in replacement for the community driver most labs already run. It matches the existing interfaces, so switching costs a single line → A C++ SDK giving the same control as a standalone library, for teams building their own stack outside ROS → Updated NVIDIA Isaac Sim assets, now supporting closed-loop kinematics That last one is HUUUGE. A 2F-85 moves through a coupled linkage, and a standard kinematic tree can't represent that properly. Closed-loop support is what lets the simulated gripper behave like the real one. The reason for the handover is load. Physical AI asks things of a gripper that conventional automation never did: higher simulation fidelity, tighter control loops, far more reporting about what is actually being held, and deployments that keep collecting data across millions of cycles. Meanwhile the tools underneath keep moving, and staying compatible is continuous work. Packages maintained in spare hours were never meant to carry that. These are also the first pieces of Contact Core, Robotiq's software layer for physical AI: simulation assets, control packages and synchronised contact-rich data interfaces for training manipulation models on Robotiq hardware and deploying them on real robots. Next comes embedded intelligence at the point of contact, shown in these demos and at IROS 2026 in Pittsburgh on 28 September, booth 427. Crediting the volunteers who carried this for years is the right way to do it 👏🏼 🔗 Check it here: robotiq.com/physical-ai ~~ ♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com