This thesis examines whether an online Monte Carlo Tree Search (MCTS)-based scheduler can improve robot path selection compared with a static First-In-First-Out (FIFO) baseline within a simulated human–robot collaborative assembly cell characterised by stochastic worker behaviour. The study focuses on reactive path planning, defined as the online selection of pre-validated robot motion paths, rather than online trajectory generation or inverse kinematics. An ABB RobotStudio virtual cell was integrated with a Python scheduling layer through an OPC UA–MQTT–Softing communication stack, using a finite RAPID path library, a stochastic two-worker model, and event-based metrics collection. The final evaluation comprised three matched FIFO/MCTS experimental pairs, with each run lasting six hours and each pair using a different workerparameter configuration. Across these scenarios, the MCTS condition produced an average of 224.3 defect-free assemblies per run, compared with 201.0 for FIFO, increasing average defect-free throughput from 33.49 to 37.38 assemblies per hour. MCTS also reduced the average robot idle fraction from 19.7% to 1.1%. The largest absolute defect-free throughput gain appeared in the asymmetric speed–quality scenario, while the smallest appeared in the symmetric high-throughput scenario. While the results are descriptive and limited to simulation-based evaluation, they suggest that MCTS has practical potential as an online scheduling mechanism when the action space is structured, the environment can be simulated, and robot execution is restricted to pre-validated paths.