TorqueAGI

TorqueAGI is a US physical-AI software company that develops foundation models for enterprise robots. Its official site describes the technology as physics-reasoning foundation models that let robots learn complex real-world tasks from minimal data and run at the edge, on the robot itself. Seed-fund F4 describes it in similar terms: the models share a common foundation that gives robots an understanding of geometry, materials and physical structure, so they can adapt across tasks and environments. The company sells B2B software to robot makers and operators rather than building its own robot hardware.

Founder. TorqueAGI is led by founder and CEO Dr. Ashutosh Saxena. He earned his PhD in computer science at Stanford under Andrew Ng, then led the Robot Learning Lab as a computer-science professor at Cornell. From 2014 to 2016 he was faculty director of RoboBrain, a joint Stanford, Cornell, Brown and UC Berkeley project that built a knowledge engine for robots. He later co-founded the smart-home company Brain of Things and the care-facility health-sensing company Caspar.AI.

Products. The website groups the models into four vertical product lines:

• TorqueFlow, for logistics and warehouses: trailer unloading, mixed-SKU sorting, kitting and packing of deformable items.
• TorqueField, for dynamic outdoor field operations: crop counting, selective harvesting, heavy-equipment operation and worksite safety.
• TorqueBuild, for assembly and manipulation: insertion, fastening and multi-step assembly.
• TorqueFocus, for inspection and quality control: defect detection, dimensional measurement, assembly checks and equipment monitoring.

Partners and customers. The TorqueAGI site names NVIDIA, John Deere, Dexterity, Berkshire Grey, GreyOrange, Canvas Robotics, JLG, Oshkosh and Pratt Miller among the companies it works with. In a May 2026 announcement the company said it is an NVIDIA Perception Partner. It added that it had been selected for John Deere's 2026 Startup Collaborator Program for the AI foundation model behind a new generation of agricultural enterprise robots. It is also working with Dexterity on truck unloading, where Dexterity's Foresight world model runs closed-loop in real time. Saxena spoke at NVIDIA GTC 2026 about running physical AI in production on NVIDIA GPUs for training and edge deployment. Earlier, in November 2024, while still in stealth, TorqueAGI signed a strategic partnership with COAST Autonomous. Under that partnership, COAST would add TorqueAGI's generative-AI agents for perception, prediction, behaviour anticipation and planning to its autonomous robots for industrial and military uses.

Funding. F4 lists TorqueAGI as a seed-stage company backed by Rackhouse Venture Capital and Trucks Venture Capital. Round sizes have not been published.

RobotToday's landscape input lists TorqueAGI under foundation models for humanoids (from the Humanoids Summit Silicon Valley 2025). Its published product lines and partners, however, are mostly wheeled, arm-based and vehicle robots in logistics, agriculture, construction equipment and factory assembly, not humanoids specifically.