A Decentralized Multi-Robot Dataset for Early Deadlock Forecasting in Smart Factory Navigation

This dataset supports research on decentralized early deadlock forecasting in multi-robot smart factory environments. It contains 100 simulated executions of two autonomous Carter robots navigating a shared factory space while visiting six predefined stations according to fixed high-level plans. Each robot executes one of five high-level plans, yielding 25 unique plan combinations, each repeated four times to capture stochastic variations in deadlock and collision occurrence. Each simulation includes: Robot trajectories (pose, velocity, orientation) at 30 Hz Navigation context tags (approaching station/ stopped at station/ stopped (unknown)) Event labels (deadlock and collision flags) Synchronized camera streams (front and back cameras per robot) Navigation maps and Nav2 parameter files Robots navigate autonomously using Isaac Sim, ROS 2 Humble, and Nav2. Logging includes timestamps and relevant sensor data, enabling experiments in early prediction, perception-aware forecasting, and neurosymbolic reasoning. The dataset is ideal for: Training models to forecast deadlocks from partial observations of a single robot Evaluating early warning and reliability trade-offs in safety-critical settings Developing interpretable neurosymbolic pipelines combining trajectories, events, and camera