The virtual worlds where robots are trained

UK startup Vsim from Cambridge developed a system for training robots in a virtual environment: walking and grasping skills are learned in minutes instead of days. The software is optimized for GPUs and runs onboard the robot, computing 20,000 scenarios per second ahead.

• Robot Freddo learned to walk and grasp a bottle in minutes
• Vsim simulator runs onboard robot and computes 20,000 scenarios
• Nvidia develops similar Isaac Sim and Cosmos world model
• Vsim has 10 engineers, founders previously worked on Isaac Sim