VARS: value-aware robotic sorting of waste streams via orienteering with time windows

The transition toward a circular economy necessitates high-quality material recovery from increasingly complex and heterogeneous waste streams. However, state-of-the-art robotic sorting algorithms primarily optimize system throughput, a proxy metric that treats all objects as equally valuable and often results in lower revenue and higher output stream contamination. This work addresses these limitations by introducing VARS (value-aware robotic sorting), a real-time decision-making framework designed to maximize total collected utility. By explicitly modeling sensor uncertainty and asymmetric contamination penalties, the framework shifts the sorting objective from simple pick-and-place (PnP) frequency to the maximization of recovered economic value and output stream purity. The proposed approach is validated through a dual-stage process involving realistic discrete-event simulations for statistical benchmarking and physical deployment on a high-speed industrial Delta robot. Results demonstrate that integrating value-based logic directly into the control loop ensures both the economic viability and operational reliability required by modern material recovery facilities.