Momentum Based Reward Design for Low Emission Traffic Signal Control

2026-05-28 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Smart Traffic Management · Depth: Expert, quick

Summary

A new Deep Reinforcement Learning (DRL) approach for adaptive traffic signal control, called the Momentum-Based Reward Function (MBRF), addresses the limitations of traditional delay and queue-based reward systems. These existing methods often lead to short-sighted or unstable policies in dynamic urban traffic conditions, contributing to congestion and environmental pollution. Proposed in a paper published on 2026-05-28, MBRF specifically encourages continuous vehicle movement rather than solely penalizing congestion. Evaluated within SUMO (Simulation of Urban MObility), the MBRF demonstrated superior performance across standard traffic metrics, including waiting time, queue length, throughput, and CO2 emissions. The results indicate that MBRF achieves better throughput-emission trade-offs and more stable learning behavior compared to both conventional DRL rewards and classical controllers like Max Pressure and LQF.

Key takeaway

For Machine Learning Engineers designing Deep Reinforcement Learning solutions for urban traffic management, consider implementing a Momentum-Based Reward Function (MBRF). Your current delay or queue-based rewards may lead to unstable policies and suboptimal environmental outcomes. Adopting MBRF can yield better throughput-emission trade-offs and more stable learning, improving overall system performance and reducing CO2 emissions in dynamic traffic environments. Evaluate MBRF against classical controllers and existing DRL rewards in simulation.

Key insights

A Momentum-Based Reward Function (MBRF) for DRL traffic control improves throughput-emission trade-offs and learning stability by prioritizing continuous vehicle movement.

Principles

• Reward design impacts DRL policy stability.
• Prioritizing flow over congestion improves outcomes.
• Adaptive control outperforms static systems.

Method

The Momentum-Based Reward Function (MBRF) for DRL traffic signal control encourages vehicle movement, contrasting with traditional delay or queue-based penalties. It's evaluated in SUMO using standard traffic and emission metrics.

In practice

• Implement MBRF in DRL traffic systems.
• Use SUMO for traffic control simulations.
• Compare DRL rewards against classical controllers.

Topics

• Deep Reinforcement Learning
• Traffic Signal Control
• Reward Function Design
• Urban Mobility Simulation
• CO2 Emissions Reduction
• SUMO

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• A decoupled diffusion planner that adapts to changing cost limits by using cost-conditioned generation for safety and reward gradients for performance — SDGD uses cost-conditioned diffusion and reward gradients for adaptive, safe, and high-performing trajectory generation.
• Chain of Uncertain Rewards with Large Language Models for Reinforcement Learning — CoUR uses LLMs and uncertainty quantification to streamline RL reward function design and evaluation.
• Application of Deep Reinforcement Learning to Event-Triggered Control for Networked Artificial Pancreas Systems — A DRL-based event-triggered controller for artificial pancreas systems improves energy efficiency by decoupling insulin dosing from update timing.
• Provable Offline Reinforcement Learning for Structured Cyclic MDPs — CycleFQI offers a modular, theoretically robust offline RL solution for cyclic MDPs with heterogeneous stages.
• Not Every Rubric Teaches Equally: Policy-Aware Rubric Rewards for RLVR — POW3R dynamically adjusts rubric reward weights during RL training to prioritize criteria that effectively teach the current policy.
• Timescale Separation Enables Deep Reinforcement Learning Control of Rotating Detonation Engine Mode Transitions — Reformulating DRL in a moving reference frame enables effective control of multi-timescale systems like RDEs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.