Abstract
This paper introduces a decentralized control-theoretic framework designed to optimize Digital Deposit Return Schemes (DDRS) within a circular economy. The proposed system addresses three critical challenges: bin congestion, incentive misalignment among supply chain actors, and static pricing. By combining stochastic load balancing via Poisson signaling, an Additive Increase Multiplicative Decrease (AIMD) algorithm for privacy-preserving reward allocation, and a Proportional-Integral (PI) feedback loop for dynamic deposit adjustment, the framework significantly improves recycling rates. Simulations using London paper cup waste data show that the system can eliminate bin-overflow waste and achieve specific recycling targets through "slow" behavioral nudging.
Introduction
The current linear economy's "take-make-waste" model is highly inefficient. For example, in the UK, approximately 2.5 billion disposable coffee cups are used annually, yet only 0.25% are recycled. Digital Deposit Return Schemes (DDRS), which assign a digital "wallet" with a refundable deposit to items, represent a promising path forward. However, their implementation is hampered by operational hurdles: uncoordinated bin usage leading to congestion, the difficulty of splitting rewards among multiple private actors (consumers, collectors, recyclers), and the lack of dynamic pricing to respond to system performance.
Methodology
The authors propose a three-tiered control strategy to manage the reverse logistics chain, modeled as a multi-layer network.
Decentralized Load Balancing
To prevent bin overflow without a central controller, bins broadcast availability signals at frequencies proportional to their remaining capacity. This creates a "competing exponential" (Poisson) process where users are naturally nudged toward emptier bins by responding to the first signal they receive.
Multi-Layer Reward Allocation
To solve the reward-splitting problem, the framework employs an Extended AIMD (Additive Increase Multiplicative Decrease) algorithm. Actors linearly increase their reward requests until the total deposit is exceeded, then decrease them based on private utility functions. This decentralized approach reaches a mathematical optimum without requiring actors to reveal their internal cost models or "price of participation."
Dynamic Deposit Pricing
A Proportional-Integral (PI) feedback loop adjusts the total deposit value based on the observed recycling rate. The controller is designed with "slow dynamics" to facilitate gradual habit formation and avoid consumer frustration from sudden price volatility.
Behavioral Modeling and Data
The framework was evaluated using high-fidelity simulations based on realistic data, including an estimated 890,000 cups consumed daily in London. A key component of the simulation is the modeling of agent behavior through concave utility functions.
As shown in Figure 4, different actors have distinct sensitivities to reward levels. The AIMD algorithm allows the system to navigate these varying thresholds to maximize total throughput while preserving the privacy of each actor's utility model.
Key Results
- Congestion Mitigation: The decentralized load-balancing algorithm reduced waste due to full bins from 13% to near 0%, matching the efficiency of a centralized deterministic controller.
- Reward Convergence: In a 20p deposit scenario, the AIMD algorithm successfully converged to an optimal split (e.g., 12p for consumers, 6p for collectors, 2p for mills).
- Target Tracking: The PI controller successfully steered the system from a near-zero recycling rate to a 50% target over 12 months, adjusting the deposit dynamically to approximately 12p.
Conclusion
This research demonstrates that "smart nudging" through decentralized control can effectively manage complex recycling networks. By prioritizing privacy and accounting for human behavioral timescales, the proposed framework offers a scalable solution for the circular economy. Future work will focus on modeling behavioral "habit lag" and validating the system through real-world pilot deployments.
Summary updated