RobOmni launched last week as a joint effort between Daimon Robotics and Galbot, two companies that have spent the past eighteen months building tactile sensor arrays for manipulation tasks in warehouses and light assembly. The benchmark suite packages twelve manipulation challenges—ranging from cable routing to fragile object handling—with corresponding tactile datasets recorded from real hardware. Each task includes ground truth force measurements, object compliance data, and success criteria measured in both task completion and damage avoidance. The platform runs on standardized gripper hardware that both companies manufacture, though the software accepts data from third-party sensors meeting the published interface specification.

The timing reflects a broader reset happening across robotics research labs and deployment teams. Vision systems have reached a practical ceiling in certain manipulation domains. A robotic arm can identify a USB cable with 99% accuracy but still fails to insert it because the system lacks feedback about contact force and angular misalignment. Autonomous mobile robots navigate flawlessly until they need to grasp a deformable bag or distinguish between a full box and an empty one of identical appearance. These failure modes all trace back to the same limitation: cameras provide rich spatial information but zero insight into physical properties like stiffness, friction, or internal structure. Tactile sensing addresses that gap, but until now the field has lacked standardized methods to evaluate competing approaches. One lab measures success by task completion rate, another by force accuracy, a third by inference speed. RobOmni attempts to establish common ground.

Daimon Robotics, based in Shenzhen, manufactures a capacitive tactile skin called SenseGrid that covers gripper fingers in a 5mm-resolution array. The company has deployed the technology in electronics assembly lines where workers previously hand-assembled ribbon cable connectors—a task that requires sensing both position and insertion resistance simultaneously. Galbot, headquartered in Hangzhou, produces a competing sensor architecture based on optical deflection measurement. Its FlexTouch sensors embed miniature cameras beneath a deformable gel surface, tracking subsurface deformation to infer contact forces and object geometry. The two companies are not natural partners—they compete directly in the contract manufacturing sensor market—but both recognized that fragmented benchmarking standards were slowing enterprise adoption. Corporate buyers evaluating tactile manipulation systems face a bewildering array of incompatible performance claims. RobOmni gives them a common reference point and gives both companies a larger addressable market if the category matures faster.

The benchmark includes tasks that expose specific failure modes in current manipulation algorithms. One scenario requires a gripper to retrieve a specific cable from a tangled bundle, a problem that appears simple but requires continuous re-planning as the robot's actions change the configuration of surrounding cables. Another test involves assembling a connector with 0.2mm tolerance while the base platform vibrates at frequencies typical of a moving fulfillment robot. A third task measures how quickly a system can classify ten different fabrics by touch alone, without visual input. Each scenario generates quantitative scores across multiple dimensions: task success rate, cycle time, force accuracy, and damage frequency. The platform also tracks computational overhead—inference time and memory footprint—to help engineers assess deployment feasibility on edge hardware. Open-source implementations of baseline algorithms are included, giving new research teams a starting point and established teams a reference for comparison.

Industry observers note that Physical AI is attracting significant venture investment even as funding for general-purpose robotics companies has contracted. Firms building tactile perception systems raised over $340 million in 2024 according to PitchBook data, more than double the prior year. Meta's FAIR lab published three papers on tactile transformers in the past six months. NVIDIA added tactile simulation capabilities to Isaac Sim in its October release. The technology is moving from research curiosity to deployment requirement, particularly in sectors where visual ambiguity creates unacceptable failure rates. Pharmaceutical packaging, which often involves translucent or reflective materials that confuse vision systems, is seeing early adoption. So is food handling, where cameras cannot distinguish ripe from overripe produce without squeezing.

What to Watch: Daimon Robotics and Galbot have committed to hosting a benchmark competition at ICRA 2025 in Atlanta this May, with cash prizes for top-performing algorithms across task categories. Both companies are negotiating with European sensor manufacturers to expand hardware compatibility—Daimon confirmed it is in talks with Germany's SynTouch and Switzerland's Xela Robotics about adding their sensor formats to the benchmark specification by June. Meanwhile, watch for Chinese electronics manufacturers, particularly Foxconn and BYD, to issue requests for proposals specifying RobOmni benchmark performance in supplier contracts, a move that would establish it as a de facto industry standard in Asia-Pacific manufacturing.