Robot manipulation with VLA models
Imitation and reinforcement learning pipelines for real-world robot manipulation built on Vision-Language-Action models.
At Araya Inc. I develop imitation learning (IL) and reinforcement learning (RL) pipelines for robot manipulation built on Vision-Language-Action (VLA) models. The work spans end-to-end training infrastructure for IL and RL on custom datasets, low-level controllers (operational-space and CBF-based), teleoperation interfaces for safe data collection, custom MuJoCo simulation environments, and the deployment infrastructure that transfers trained policies onto physical hardware.