Phase II Amount
$2,499,994
GelSight, Inc. proposes GelSight Nano, a sub-15 mm-thick, image-based tactile fingertip that equips robotic grippers and humanoid hands with human-level touch. Todays industrial robots cannot reliably feel slip, force, or contact geometry; competing optical fingertips remain bulky, slow, and not ruggedized for industrial environments. GelSight Nano overcomes those limits by uniting a folded-optics, wafer-level-camera stack with a robust elastomeric sensor that sustains more than 50,000 grasps while streaming 50 fps 3D shape and force maps across a 15 mm × 15 mm field of view. The 21-month Phase II program will first compress the optical path and mechanical housing to achieve a form factor less than 15 mm thick, then implement APIs to deliver rich tactile data at 50 Hz. Durability trials on an automated wear rig will down-select elastomer formulations to keep opacity loss under five percent after 50,000 shear cycles, while a simulation engine will generate domain-randomized tactile image data so reinforcement-learning policies can be trained in simulation and transferred to hardware to reduce the training burden on real robotic systems. Prototype fingertips will be integrated on multi-fingered graspers before undergoing a TRL-6 field demonstration at Warner Robins Air Logistics Complex. A structured customer-discovery effort will refine value proposition, key applications, channel strategy, pricing, and a five-year revenue forecast. By merging a compact form factor, long-life gels, high-throughput electronics, and AI-ready data streams, this effort positions GelSight as the leader for robotic touch for DoD maintenance robots and the rapidly expanding humanoid market, enabling dexterous automation across aerospace, defense, and advanced manufacturing. ------------- Automated dexterous manipulation is critical for enhancing efficiency and precision in aerospace manufacturing and Air Force maintenance operations, where current manual methods are slow, error-prone, and costly. Robotic systems with tactile sensing can significantly improve safety, cost-effectiveness, and readiness. However, existing tactile sensors are too bulky (2530 mm), lack durability (under 10,000 uses), and have slow data capture rates (25 fps), making them unsuitable for real-world robotic integration. To overcome these challenges, a compact (under 20 mm), high-speed (over 50 fps) tactile sensor will be developed and integrated into multi-fingered robotic hands. First, the tactile sensor will be designed, focusing on optical and material improvements for enhanced durability and sensitivity. The sensors will then be integrated into robotic platforms for dexterous manipulation. Using teleoperation, skilled operators will guide the robots through complex tasks, generating high-quality tactile data. This dataset will be ingested by machine learning models aimed at improving manipulation skills through simulation and real-world feedback. Simulations will model the sensors behavior to enable advanced capabilities like slip detection and adaptive force control. The goal is to develop a semi- to fully autonomous robotic system capable of precise, tactile-guided manipulation for logistics, assembly, and inspection tasks.
