
SAN JOSE, Calif., March 18, 2026 /PRNewswire/ -- Dexterity.ai, the leader in industrial robotics and Physical AI, today detailed how its production-proven world model, Foresight, is unlocking new levels of performance on NVIDIA hardware as the company scales in enterprise environments.
Foresight is Dexterity's world model, built from the experience of over 100 million autonomous actions in production environments. It gives robots real-time understanding of the physical world: reasoning about object geometry, contact dynamics, weight distributions, stability, and more, before every action. Every decision Foresight makes is fully interpretable and safe.
Before any action or motion planning can begin, Foresight must build a complete picture of the scene from raw multi-modal sensor data. Dexterity's engineering team redesigned the perception pipeline to take full advantage of NVIDIA L4 GPUs and TensorRT, reducing it from 1,508ms to 90ms per cycle, a 17x speedup across three stages: object decomposition, geometric reconstruction, and physics synthesis.
The most significant gain was not speed alone but data utilization. "We went from processing 3% of available data to all of it; that is 32x more data per cycle that gives the system a far richer understanding of re every scene, and decision cycles below 400ms in highly complex applications such as truck loading," said Robert Sun, Founding Engineer at Dexterity.
Recognized by NVIDIA at GTC 2026
NVIDIA highlighted the collaboration at GTC 2026, naming Dexterity among the Physical AI pioneers taking robotics to the real world.
"Physical AI has been a research topic for decades. What's changed is that it now works extremely well in production," said Samir Menon, CEO and co-founder of Dexterity. "Foresight is running at enterprise scale today, handling complex manipulation tasks in complex customer environments. When the technology ecosystem led by NVIDIA and enterprise customers are both moving in this direction, this is just the beginning of what Foresight will do."
Showcased at FedEx Investor Day
Dexterity showcased its Foresight-powered truck loading at FedEx's 2026 Investor Day in Memphis. Attendees watched Dexterity's dual-armed superhumanoid robot, Mech, autonomously load a trailer, receiving a stream of randomly shaped packages and constructing tightly packed, stable walls with real-time AI decision-making using Foresight.
During the Investor Day presentation, FedEx discussed its intent to scale trailer loading and unloading across several US hubs over the next few years. FedEx leadership subsequently recognized Dexterity's role in advancing trailer loading operations at the event. Dexterity and FedEx had previously unveiled the trailer loader in 2023.
About Dexterity
Dexterity.ai builds the intelligence layer that enables robots to handle the full complexity of real-world production environments. The company's world model, Foresight, combines multi-modal sensing, physical reasoning, and predictive planning to enable reliable robotic manipulation at enterprise scale. For more information, visit dexterity.ai.
Media Contact: Claudia Weise, [email protected]
SOURCE Dexterity AI
Share this article