Sim-to-Real Transfer
A core technical challenge in robot learning that involves transferring policies trained in simulated environments to real physical robots. It requires overcoming the gap between simulation and reality, including differences in friction, noise, and latency.
Timeline (last 90 days)
A Reality Gap exists between manually constructed simulators and the real world, causing policy performance to degrade in reality
A 2023 study by a Berkeley team demonstrated that bipedal robot policies trained through large-scale domain randomization can achieve zero-shot transfer to real hardware.
All Facts (2)
A Reality Gap exists between manually constructed simulators and the real world, causing policy performance to degrade in reality
50%UnverifiedA 2023 study by a Berkeley team demonstrated that bipedal robot policies trained through large-scale domain randomization can achieve zero-shot transfer to real hardware.
50%