Agentic Dora Agent Driven Robot Control

Dora is a powerful robotics dataflow framework, but robot behaviors require fixed YAML graphs—no natural language interface, runtime error recovery, or high-level goal decomposition. This project builds an Octos agent bridge node for Dora, enabling natural language control of a simulated UR5e robot (e.g., "pick up the red ball and place it on the green plate"). The agent decomposes commands into DOT-graph pipelines, executing via dora transport tools, interfacing with dora-moveit nodes in MuJoCo. A prototype demonstrates pick-and-place with full step completion (vs partial completion with LLM-only reasoning). Deliverables: MuJoCo simulation, Octos bridge node, agent-callable tools, DOT-graph pipeline, mock provider, test suite, documentation. Stretch: Rust-native bridge, vision-based grasping, messaging channel integration