Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Coordination of Emergent Demand Changes via Value-Based Negotiation for Supply Chain Management (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes value-based automated negotiation between RL supply-chain agents (PPO-trained, value function used as negotiation utility) to handle emergent demand changes, avoiding inventory shortages.
Problem
Reinforcement learning for supply chain management assumes centralized hierarchical coordination and cannot adapt to unexpected demand changes through inter-agent negotiation.
Approach
Uses the RL agent's value function, via multiple internal simulations, as the utility function in an alternating-offers negotiation; negotiation is triggered when the gap between learned value and simulated value exceeds a threshold, and agents negotiate shipping plans with linear concession strategies.
Key results
- PrePV-Nego achieved an achievement rate of 1.00 for both supplier and customer in all three demand-change scenarios.
- Zero-Shot dropped to 0.43-0.97 across scenarios, and Base-Stock to 0.67-1.00.
- PrePV-Nego shipped on average 33% more and held 52% more inventory than Zero-Shot.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use value-based negotiation instead of Zero-Shot without retraining | PrePV-Nego vs Zero-Shot · Decrease(small), supplier (S) and customer (C) | achievement rate: 0.73 -> 1.00; 0.97 -> 1.00 | Helped |
| Use value-based negotiation instead of Zero-Shot without retraining | PrePV-Nego vs Zero-Shot · Increase(small), supplier (S) and customer (C) | achievement rate: 0.80 -> 1.00; 0.87 -> 1.00 | Helped |
| Use value-based negotiation instead of Zero-Shot without retraining | PrePV-Nego vs Zero-Shot · Increase(large), supplier (S) and customer (C) | achievement rate: 0.63 -> 1.00; 0.43 -> 1.00 | Helped |
| Use value-based negotiation instead of Base-Stock | PrePV-Nego vs Base-Stock · Increase(small), supplier (S) and customer (C) | achievement rate: 0.90 -> 1.00; 0.83 -> 1.00 | Helped |
| Use value-based negotiation instead of Base-Stock | PrePV-Nego vs Base-Stock · Increase(large), supplier (S) and customer (C) | achievement rate: 0.80 -> 1.00; 0.67 -> 1.00 | Helped |
| Use value-based negotiation instead of Base-Stock | PrePV-Nego vs Base-Stock · Decrease(small), supplier (S) and customer (C) | achievement rate: 1.00 -> 1.00; 1.00 -> 1.00 | No clear effect |
Limitations
- Evaluated only in a simplified four-layer serial simulation with a single shared demand trajectory.
- Student abstract with limited experiments and only three demand-change scenarios.
Experimental setup
Datasets: simulated four-layer serial supply chain with stochastic terminal customer demand
Benchmarks: custom supply chain simulation scenarios (Decrease small, Increase small, Increase large)
Metrics: achievement rate
Baselines: Zero-Shot Base-Stock Ideal
Builds on: proximal policy optimization alternating offers protocol
Closest papers by meaning
- OFCOURSE: A Multi-Agent Reinforcement Learning Environment for Order Fulfillment (NeurIPS 2023 · similarity 0.84)
- Online Learning in the Repeated Mediated Newsvendor Problem (NeurIPS 2025 · similarity 0.83)
- Learning in Online Principal-Agent Interactions: The Power of Menus (AAAI 2024 · similarity 0.81)
- SrSv: Integrating Sequential Rollouts with Sequential Value Estimation for Multi-agent Reinforcement Learning (AAAI 2025 · similarity 0.80)
- Contracting with a Learning Agent (NeurIPS 2024 · similarity 0.80)
- Explicit Exploration for High-Welfare Equilibria in Game-Theoretic Multiagent Reinforcement Learning (ICML 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.