RoboControl

RoboControl is an AI mission-control system for autonomous rover fleets. It turns natural-language or voice commands into safety-validated robot operations, giving an operator one command layer for planning, robot selection, execution, telemetry monitoring, recovery, and audit logging. This is not just a dashboard or chatbot: RoboControl sends executable movement commands through a rover bridge, so a simulated, Gazebo/ROS-connected, or hardware-adapted rover can actually move under the same safety-gated runtime. The system follows a real agent lifecycle: planning → safety validation → execution → telemetry monitoring → recovery → memory update. When an operator says “survey the area, avoid hazards, and report status,” RoboControl converts that goal into a structured action plan, selects the best rover based on capabilities and health, checks the action against safety policy, executes drive or mission commands, watches telemetry, and generates voice briefings or hazard alerts. RoboControl combines a FastAPI backend, Next.js fleet dashboard, robot registry, runtime orchestrator, rover bridge, ROS/Gazebo deployment path, ElevenLabs voice interface, and an AMD/Qwen planner layer. The Qwen planner is designed for LoRA specialization on AMD Developer Cloud so rover mission data can train a compact model for JSON action planning, robot routing, and safety-aware decisions. The repo includes dataset generation, fine-tuning, evaluation, and vLLM serving scripts for AMD ROCm environments. Unlike generic chat agents, RoboControl is built for action. Every decision can include operation ID, robot ID, agent role, lifecycle stage, severity, and safety outcome. Manual drive commands also pass through safety validation. ElevenLabs adds voice commands, mission briefings, spoken confirmations, and telemetry-specific hazard alerts, with graceful fallback when no voice key is configured.