Multi-agent elevator dispatch with QR-DQN and CTDE (Centralised Training, Decentralised Execution) in PyTorch. Three cooperative elevators trained with dueling networks, prioritised replay, and a centralised critic — beating ETA-Dispatch by 27% wait and 9.6% energy across 3,500 episodes on a non-uniform 10-floor traffic environment.
python reinforcement-learning deep-reinforcement-learning pytorch deep-q-network prioritized-replay quantile-regression dueling-dqn multi-agent-reinforcement-learning qr-dqn marl elevator-scheduling cooperative-ai decentralized-execution ctde
-
Updated
Apr 26, 2026 - Jupyter Notebook