Towards Autonomous Micromobility through Scalable Urban Simulation

TL;DR

This work introduces URBAN-SIM, a high-performance urban simulation platform, and URBAN-BENCH, a comprehensive suite of tasks for autonomous micromobility. URBAN-SIM combines Hierarchical Urban Generation, Interactive Dynamics Generation, and Asynchronous Scene Sampling to deliver infinite, realistic urban scenes with GPU-accelerated training. URBAN-BENCH spans Urban Locomotion, Urban Navigation, and Urban Traverse, evaluated across four robot embodiments using PPO-based learning; results reveal embodiment-specific strengths and the value of human–AI shared autonomy for long-horizon tasks. The framework demonstrates strong scalability and lays groundwork for sim-to-real deployment and open-world learning, aiming to accelerate advancements in embodied AI for safe, efficient urban mobility.

Abstract

Micromobility, which utilizes lightweight mobile machines moving in urban public spaces, such as delivery robots and mobility scooters, emerges as a promising alternative to vehicular mobility. Current micromobility depends mostly on human manual operation (in-person or remote control), which raises safety and efficiency concerns when navigating busy urban environments full of unpredictable obstacles and pedestrians. Assisting humans with AI agents in maneuvering micromobility devices presents a viable solution for enhancing safety and efficiency. In this work, we present a scalable urban simulation solution to advance autonomous micromobility. First, we build URBAN-SIM - a high-performance robot learning platform for large-scale training of embodied agents in interactive urban scenes. URBAN-SIM contains three critical modules: Hierarchical Urban Generation pipeline, Interactive Dynamics Generation strategy, and Asynchronous Scene Sampling scheme, to improve the diversity, realism, and efficiency of robot learning in simulation. Then, we propose URBAN-BENCH - a suite of essential tasks and benchmarks to gauge various capabilities of the AI agents in achieving autonomous micromobility. URBAN-BENCH includes eight tasks based on three core skills of the agents: Urban Locomotion, Urban Navigation, and Urban Traverse. We evaluate four robots with heterogeneous embodiments, such as the wheeled and legged robots, across these tasks. Experiments on diverse terrains and urban structures reveal each robot's strengths and limitations.

Figures (22)

• Figure 1: Autonomous micromobility. In public urban spaces, various mobile machines (circular images) are essential for short-distance travel. However, urban environments are complex and contain varied terrain and challenging situations (rectangular images). To bridge this gap, we present a scalable urban simulation solution to advance autonomous micromobility. Images are from our Urban-Tra-City data.
• Figure 2: URBAN-SIM: a robot learning platform for autonomous micromobility. (a) Hierarchical Urban Generation. It generates an infinite number of diverse scenes through four progressive stages. (b) Interactive Dynamics Generation. GPU-based generation of realistic agent-scene and agent-agent interactions on the fly. (c) Asynchronous Scene Sampling. An asynchronous sampling scheme to enable high-efficiency training on varied scenes with rich context information.
• Figure 3: Scene sampling diagram. (Left) Assets Cache that stores all assets in urban scenes. (Right) With a random sampling of assets, parallel environments can be constructed on GPU.
• Figure 4: URBAN-BENCH: a suite of essential tasks for autonomous micromobility. Simulation environments of eight essential tasks of (a) Urban Locomotion, (b) Urban Navigation, and (c) Urban Traverse.
• Figure 5: Emerging behaviors. The results of evaluating different robots in the same environment. After training in diverse urban scenes, robots with distinct structures have developed their unique movement skills.