Dexterity has introduced Foresight, a world model designed to support robotic systems performing physical tasks such as loading freight into trucks. The technology is part of the company’s software stack for industrial robotics and is intended to enable robots to perceive and simulate physical environments in real time while making placement decisions for packages.
Foresight is described as a physics-based representation of the physical environment that allows robots to interpret surroundings and plan actions. The system supports robotic manipulation tasks that require continuous evaluation of spatial relationships and constraints. According to the company, the model operates in real time and maintains a transactional representation of the environment that can be updated as objects move or conditions change.
In truck-loading applications, Foresight powers Dexterity’s dual-armed robot, known as Mech. The system uses what the company describes as a four-dimensional packing agent that evaluates three spatial dimensions along with time to determine where to place each package on a developing wall of freight. The model evaluates potential placements for individual boxes while accounting for density, structural stability, reachability and coordination between the robot’s two arms.
Dexterity said the system may evaluate hundreds of possible placements for a single package and produce a decision in under 400 milliseconds. The model also predicts how each placement affects the structure of the freight load as additional packages are added.
The world model operates within Dexterity’s broader robotics architecture, which links perception, planning and motion components. The framework coordinates agents responsible for sensing the environment, making decisions and controlling robotic movement. According to the company, the system is designed to allow operators to view the reasoning behind individual decisions made by the robot.
Dexterity stated that its robotics software stack can operate across multiple applications and hardware configurations. The company said the system has been used in applications including truck loading, parcel singulation, palletizing and depalletizing. It added that the model has been trained using data from more than 100 million autonomous robotic actions in operational environments.
Samir Menon, founder and chief executive of Dexterity, said the system enables robots to make placement decisions by predicting how individual actions affect the overall freight structure.
Dexterity also introduced the Foresight API Challenge, a competition for student teams to develop packing agents using the system. Participants compete on a public leaderboard, with prize funding totaling up to $50,000. The competition requires teams to create their own models of physical behavior rather than using a provided simulator.
Dexterity was founded in 2017 in Stanford University’s robotics laboratory and develops robotics systems intended for warehouse and logistics operations. Its products combine robotic hardware with software models designed to coordinate perception, planning and motion for automated handling tasks.
