Repository navigation
PPO Elevator v1.0.0
First stable release of the Elevator-AI project.
This release includes a complete custom Gymnasium simulation environment for a 20-floor, 4-elevator building with probabilistic traffic patterns and a full two-phase passenger model. A classical Destination Dispatching baseline and a PPO-trained reinforcement learning agent are both included, along with evaluation scripts for reproducible benchmarking.
Key result: The PPO agent achieves a mean average reward of +0.14 vs -0.67 for the classical baseline — driven by an 84.5% reduction in average passenger wait time (93 vs 601 steps) over 1,000 evaluation episodes.
Training: 5,000,000 steps on CPU, trained on a MacBook Pro M2 Pro in approximately 25 minutes. No GPU required.
Realistic elevator kinematics (speed, door cycles, dwell time) are in active development for v1.1.0.