Planning and Decision Making

Planning and decision making encompass the processes by which individuals or organisations identify objectives, generate and assess alternative courses of action, and select strategies that best satisfy competing requirements under uncertainty. At their heart lie models of causality and purpose: causal analysis reveals how actions lead to outcomes, while teleological reasoning links those outcomes to stakeholder goals. Techniques span from precise mathematical programming—single­ and multi­objective optimisation—to heuristic, rule­based and learning methods that handle complexity, dynamism and partial information. Modern decision support systems integrate data management, model‐driven analysis and interactive interfaces to guide users toward optimal, Pareto‐efficient or satisficing solutions, balancing cost, risk and strategic priorities.

Research from Nature Portfolio

Recent work has demonstrated a detector‐free, citywide traffic signal re‐timing system that uses only sparse vehicle trajectories rather than fixed detectors. By constructing a probabilistic time–space diagram under Newellian coordinates, researchers reconstructed network‐wide traffic states and formulated optimisation algorithms to update signal parameters. In a real‐world trial, this approach reduced intersection delay by up to 20 % and cuts stops by 30 %, illustrating how scalable, data‐driven signal control can deliver near‐optimal performance under stochastic flows. Another study proposed a decentralised multi‐agent RL framework for traffic lights, featuring novel message‐synchronisation and queue/waiting‐time reward calculations. Each intersection agent exchanges concise traffic descriptors and computes rewards combining queue length and delays, yielding efficient coordination without requiring extensive communication overhead and outperforming conventional single‐agent methods in congestion reduction.

Research from all publishers

A systematic review of RL applications to urban network signal control synthesised over 160 studies and highlighted the transition from early Q‐learning to deep actor‐critic methods. It underscored unified evaluation metrics, the rise of open‐source microsimulators and the challenges of real‐world deployment, including reward design, safe exploration and state representation. Complementing this, a multi‐agent deep RL approach for connected autonomous vehicles at intersections employed curriculum self‐play to train cooperative controllers that replace fixed signals. In simulations it cut travel time by over 50 % and congestion‐related delay by 95 %, demonstrating the potential of model‐based learning for adaptive corridor management. A broader DRL review detailed architectures—from deep Q‐networks to proximal policy optimisation—surveyed popular simulation environments and identified future directions such as multi‐objective control and integration with traffic demand forecasting.

Planning and Decision Making publication trend

The graph below shows the total number of articles in planning and decision making across all publications each year (not limited to Nature Index journals).

Technical terms

Teleology: Study of purpose or goal‐directed processes, distinguishing intent‐based actions from purely causal phenomena.

Pareto optimality: A state in which no objective can be improved without degrading at least one other objective, used in multi‐objective optimisation.

Reinforcement learning (RL): A paradigm in which agents learn decision policies through trial‐and‐error interactions by maximising cumulative rewards.

Multi‐agent system: An environment where multiple learning agents interact, coordinate or compete to achieve individual or collective objectives.

Probabilistic time–space diagram: A stochastic model linking vehicle trajectory data to queue dynamics, employed to infer traffic states for control optimisation.

References

1. Traffic light optimization with low penetration rate vehicle trajectory data. Nature Communications (2024).
2. A traffic light control method based on multi-agent deep reinforcement learning algorithm. Scientific Reports (2023).
3. Reinforcement learning in urban network traffic signal control: A systematic literature review. Expert Systems with Applications (2022).
4. Multi-Agent Deep Reinforcement Learning to Manage Connected Autonomous Vehicles at Tomorrow's Intersections. IEEE Transactions on Vehicular Technology (2022).
5. Deep Reinforcement Learning for Traffic Signal Control: A Review. IEEE Access (2020).