CC
Cobot Claw: Agentic Control for Collaborative Robots
Feb 2026 - Present
Multi-agent AI system for natural-language control of UR collaborative robot arms and end-effectors, designed for smart factory and industrial automation environments.
Problem Statement
- Industrial robot operation still depends heavily on specialist programming workflows, slowing adoption on the factory floor.
- End-effectors such as grippers, vacuum pumps, and custom tools require flexible integration without rebuilding the control stack each time.
- Safe autonomy demands continuous monitoring and intervention rather than one-shot command execution.
What I Led
- Architected a multi-agent, LLM-powered control system coordinated via MCP servers, enabling hardware-agnostic tool discovery across robotic arms and end-effectors.
- Designed a safety-first autonomy layer with a dedicated monitoring agent that continuously observes robot state and intervenes in real time when execution drifts from safe conditions.
- Enabled deployment across local and cloud inference backends, including DGX Spark and hosted APIs, with a one-line configuration switch.
Deliverables
- Multi-agent natural-language robot control architecture
- MCP-based tool discovery for diverse end-effectors
- Real-time robot monitoring and intervention workflow
Tech Stack
PydanticAILlamaIndexROS 2FastMCPOllamaOpenAI SDKAnthropic SDK
Industry Relevance
Demonstrates a practical path for manufacturers to reduce robotics integration friction while keeping safety, observability, and deployment flexibility central to production operations.