Campus-Scale Waste Collection Challenge: A Unified Framework for Robotic Perception, Navigation, and Manipulation

This paper introduces the Campus-Scale Waste Collection Challenge, a simulation-based benchmark for autonomous mobile manipulation in outdoor environments with perceptual uncertainty. Built in NVIDIA Isaac Sim as a digital twin of a real university campus, the environment includes diverse terrain, dynamic obstacles, and variable lighting. Participants must detect, localize, and classify waste using partial CCTV observations, and perform full pick-and-place tasks with wheeled manipulators. A ROS 2-based API enables integration of perception, planning, and control modules. The challenge evaluates performance across perception, navigation, and manipulation stages using metrics such as accuracy, success rate, and execution time. A baseline implementation, dataset, and imitation learning pipeline are provided to support reproducibility and development. This testbed promotes robust, scalable solutions for real-world service robotics.