VARS introduces a real-time decision-making architecture that reframes robotic waste sorting as an expected utility maximization problem to prioritize high-value, low-contamination items over purely throughput-driven picks.

Physical experiments align closely with simulations, with differences typically under 1%, and the planner tends to target high-value items while bypassing ambiguous ones to preserve purity and prevent downcycling.

Validation combines a discrete-event simulation digital twin (Salabim) and physical testing on an industrial Delta robot with a four-bin sorting cell, a five-class polymer stream, and varying arrival rates to compare against throughput-based baselines.

A comprehensive loss-penalty and utility structure uses posterior class probabilities from sensing (RGB-D and hyperspectral data) mapped through a utility matrix that imposes asymmetric costs for misclassification to reflect downstream impact.

The study integrates sensing, probabilistic reasoning, and combinatorial optimization to improve economic and environmental outcomes in automated recycling.

Limitations include assumptions of unclustered, mildly overlapped items and single-material plastics; future work aims to handle cluttered 3D environments, incorporate volume/mass and market pricing into utilities, and expand to cooperative multi-robot sorting.

Key findings show value-aware approaches maintain higher recovered utility and output stream purity under higher load, even with perception degradation, and Iterated Local Search can outperform exact solvers in highly loaded scenarios within practical latency.

The framework models sorting as an Orienteering Problem with Time Windows, incorporating time-dependent feasibility due to continuous conveyor motion and uncertainty-aware economic utility from material classification.

Decision-making uses both greedy rules (Maximum Value, Value-to-Time Ratio) and sequence-level optimization via exact Mixed-Integer Programming and an Iterated Local Search metaheuristic, achieving near-optimal sequencing within sub-second budgets.