| Project ID |
BITS-SRIP/D85787/2026 |
| Project Title |
Vehicle composition aware delay and energy minimization signal control |
| Project Description |
Efficient signal control at urban intersections is critical for reducing congestion, travel delay, and energy consumption. While numerous signal optimization algorithms, such as Max Pressure (MP) and Reinforcement Learning (RL) have been proposed, most are designed for homogeneous, lane-disciplined traffic conditions typical of developed countries. In contrast, traffic in developing nations such as India is highly heterogeneous and less lane-disciplined, consisting of a diverse mix of vehicle types (two-wheelers, cars, auto-rickshaws, buses, and trucks) that exhibit non-FIFO and limited lane discipline behavior. Existing control systems in such contexts largely rely on fixed-time, actuated control or limited composite strategies like CoSiCoSt, which do not adequately capture the dynamics of heterogeneous traffic.
This study proposes a hybrid, adaptive signal control framework applicable for heterogeneous, less lane-disciplined (HLLD) traffic environments. The proposed approach explicitly incorporates vehicle composition into signal timing decisions, enabling phase-wise green time optimization that accounts for both person-delay and vehicle-type-specific energy consumption. By jointly minimizing delay and energy consumption objectives that vary significantly with traffic composition, the framework aims to improve operational efficiency and sustainability. The proposed vehicle-composition-aware control strategy demonstrates strong potential for reducing queues, delays, fuel consumption and emissions in real-world urban intersections under heterogeneous traffic conditions.
The proposed framework integrates microscopic traffic simulation, trajectory optimization and reinforcement learning to capture the complex dynamics of real-world heterogeneous traffic. For a given signal setting, microscopic simulations are first executed to obtain detailed vehicle trajectories from an upstream reference point to a downstream point of the signalized intersection. The simulation outputs include individual vehicle entry and exit times, vehicle type, delay, and speed-acceleration profiles. Using these outputs, queue formation and dissipation dynamics are extracted by calibrating an Intelligent Driving Model (IDM) suitable for heterogeneous traffic. Based on the inferred queue dynamics, original vehicle trajectories are transformed into energy-efficient trajectory profiles consisting of three distinct motion phases: acceleration/deceleration, constant cruising and final deceleration/acceleration. For fixed signal timings and fixed entry-exit times, these modified trajectories are generated such that the total energy consumption for each vehicle between the upstream and downstream locations is minimized.
In the second stage, optimal signal timings are computed with the primary objective of minimizing person-delay. Given these optimal signal timings, total energy consumption is evaluated for both the original (using fixed time signal setting) and the modified vehicle trajectories (using optimal signal setting). If the energy consumption associated with the modified trajectories is lower than that of the original trajectories, the person-delay-optimal signal timings and the corresponding modified trajectories are retained and implemented for the current signal cycle.
However, if the modified trajectories result in higher energy consumption than the original trajectories, the framework switches to an alternative control strategy. In this case, vehicle arrival times at the stop line are recomputed using the original trajectories, and new optimal signal timings are derived with the objective of maximizing person throughput at the intersection. This adaptive decision-making process enables dynamic trade-offs between delay minimization and energy efficiency, depending on prevailing traffic composition and flow conditions.
To operationalize this multi-objective and adaptive control logic under real-world heterogeneous traffic conditions, the study proposes the use of a multi-objective Reinforcement Learning (RL) framework implemented via the VISSIM–COM interface. The RL agent learns optimal phase-wise green allocations by observing traffic states characterized by vehicle composition, queue dynamics, and arrival patterns, thereby enabling robust and scalable signal control for HLLD traffic environments.
Expected Outcomes:
(i) Development of a vehicle-composition-aware adaptive signal control framework for heterogeneous, less lane-disciplined traffic.
(ii) Reduction in person-delay and queue lengths through phase-wise green time optimization.
(iii) Minimization of vehicle-type-specific energy consumption using energy-efficient trajectory planning.
(iv) Demonstration of trade-offs between delay, throughput, and energy objectives under varying traffic compositions.
(v) Practical guidelines for deploying intelligent signal control in developing-country traffic conditions.
(vi) Direct contribution to Sustainable Development Goals. |
| Project Discipline |
Civil Engineering |
| Faculty Name |
Dr. Maripini Himabindu |
| Department |
Department of Civil Engineering |
