World Generation
Describe a building in plain text and Arena generates the floorplan, furnishes it, and turns it into a world every simulator can load, or author multi-floor worlds with doors and elevators by hand.
Platform Modules
Arena is an open ROS 2 platform for building, running, and comparing robot navigation in realistic indoor spaces full of people, on the simulator of your choice.
Pedestrian Simulation
Crowds of animated people (staff, visitors, elderly walkers, and wheelchair users) walk, queue, gather, talk, and gesture, and they react to the robot.
Scenario Generation
Arena fills any world with everyday crowds that fit each room's purpose, for example people at work, on a break, or on a night shift. It then turns plain-language prompts such as "an old man crosses the robot's path", "a power cut leaves the floor in the dark", or "the fire alarm goes off" into ready-to-run test cases. The generated cases bundle straight into a benchmark.
Lighting and Sound
Worlds declare their lights, which every simulator renders and which can dim or fail mid-run. A simulated soundscape lets robots hear footsteps, speech, and alarms through their own microphones.
Training
Train deep reinforcement learning navigation agents in many parallel simulated environments, then deploy the trained policy as a planner in the same worlds.
Benchmarking
Arena runs planners through the same worlds, crowds, and tasks, judges every episode, and produces a report that compares them on success, efficiency, safety, social compliance, and energy. Large comparisons can be split across parallel isolated copies of the platform to finish faster.
