Microsoft adds remote AI inference tools for robots

Microsoft Research has added remote AI inference support to its Physical AI Toolchain, allowing developers to distribute robot workloads across onboard computers, nearby servers and cloud GPUs. The release includes example projects for SO-101 and UR10e robots and uses Kubernetes to package and place the workloads. In Microsoft's tests of mobile robot tasks, shifting computation away from onboard GPUs improved response time and accuracy while reducing battery drain. The approach can also let robots use larger models than their onboard hardware can hold. Its benefit depends on network latency, bandwidth and access to remote GPUs, which can constrain robots working in unpredictable environments.