robotics research

The robotics research tag follows work on improving robot learning, manipulation, and real-world reliability through simulation, foundation models, and iterative experimentation. Covered reporting includes Microsoft Research’s object-centric residual reinforcement learning approach, which adds a lightweight corrective policy to a frozen vision-language-action model and tests zero-shot gains across five manipulation tasks. It also includes Nvidia’s ENPIRE framework, where AI coding agents run hardware experiments, check results, revise policy code, and pursue tasks such as installing a GPU into a motherboard. Together, these stories examine how software, simulation, and physical testing are being combined to move robotics beyond compelling demonstrations toward more dependable execution.