Towards Autonomous Micromobility through Scalable Urban Simulation (2505.00690v1)

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.

Autonomous Micromobility in Urban Environments: Simulation and Benchmarking

The paper under review presents a significant contribution to the domain of autonomous micromobility through the development of a scalable urban simulation platform, URBAN-SIM, coupled with a benchmarking suite, URBAN-BENCH. The overarching goal of this work is to facilitate the advancement of autonomous micromobility, particularly focusing on the operation of lightweight devices such as delivery robots and mobility scooters in complex urban terrains.

Key Developments in Simulation and Benchmarking

The paper introduces URBAN-SIM, a high-performance robot learning environment designed to enable large-scale training of AI agents in dynamic and diverse urban environments. This platform is characterized by three core components: a Hierarchical Urban Generation pipeline, an Interactive Dynamics Generation strategy, and an Asynchronous Scene Sampling scheme. These elements collectively enhance the diversity, realism, and training efficiency of simulated environments.

• Hierarchical Urban Generation: This component allows the procedural generation of diverse urban layouts to train agents across varying terrains and urban configurations—ranging from street blocks to complex urban settings—thereby ensuring robust generalization capabilities of models.
• Interactive Dynamics Generation: By leveraging real-time multithreaded simulations executed on GPUs, this strategy ensures dynamic interactions between multiple agents, enhancing the realism of training scenarios with realistic simulations of pedestrian and vehicular movement.
• Asynchronous Scene Sampling: This method allows the parallel sampling of numerous unique training environments in a single pass, significantly improving the efficiency of large-scale training and enabling substantial scalability.

Alongside, the paper presents URBAN-BENCH, a comprehensive set of tasks and performance benchmarks. This suite provides eight tasks classified under three main skill areas: Urban Locomotion, Urban Navigation, and an ambitious Urban Traverse task. These categories encompass critical capabilities required for autonomous micromobility, such as navigating static and dynamic obstacles, and scaling varied urban terrains.

Experimental Findings and Analysis

The paper conducts detailed benchmarks across multiple robotic platforms, including wheeled, quadruped, and humanoid robots, to evaluate various capabilities. Notably, the paper underscores that:

• Quadruped robots demonstrate superior traversal capabilities, particularly on complex terrains like stairs and rough surfaces, confirming the significance of specialized mechanical structures in handling intricate urban environments.
• In contrast, wheeled robots show optimal performance in pathfinding tasks devoid of obstacles, indicating their efficiency on flat, predictable surfaces.
• In Urban Navigation tasks, humanoid robots perform well in dynamically populated scenarios, reflecting their adaptability in maneuvering through crowded spaces.

Additionally, the research investigates a Human-AI Shared Autonomy approach, addressing kilometer-scale Urban Traverse. Through this approach, seamless integration of human decision-making and autonomous control is achieved, enabling flexible and adaptive navigation strategies that responsive to real-time challenges in urban landscapes.

Implications and Future Work

The implications of this research extend across multiple domains:

• Collaborative Urban Robotics: The development of URBAN-SIM and URBAN-BENCH provides a scalable framework to explore collaborative multi-agent systems, essential for future urban mobility solutions.
• Urban Planning and Accessibility: By enabling simulations of urban environments, planners can optimize infrastructure for increased flexibility and accessibility, reducing dependency on traditional vehicular systems for short-distance urban travel.
• AI Deployment and Safety: Insights from these benchmarks can guide the deployment of safe and efficient autonomous navigation systems, aligning with emergent regulatory and operational standards in urban contexts.

Looking forward, this paper envisions further enhancements targeting real-world deployments and improvements in sim-to-real transfer capabilities. Emphasizing scalability, extensibility, and community-driven development, this research aims to provide a comprehensive ecosystem that supports ongoing advancements in embodied AI and autonomous micromobility. The introduction of more diverse tasks, real data integration, and broader robotic capabilities are key directions for future research, promising to expand the utility and impact of this work in the dynamic field of urban robotics.