Simulated ‘Plug-and-Play’ Control Layer Cuts Fuel Use in Gasoline Vehicle Platoons

A simulation study published in the journal Discover Vehicles reports that a supervisory control layer, bolted onto the outside of ordinary vehicle controllers without touching their internal settings, can measurably reduce fuel consumption and nitrogen oxide emissions in platoons of gasoline-powered autonomous vehicles. The framework, developed by Behrouz Samieiyan and Anjali Awasthi, combines a neuro-fuzzy engine model, model predictive control, and a quantum-inspired long short-term memory network that broadcasts a ten-second forecast of the lead vehicle’s behavior to every follower at once. Across four standardized highway driving cycles and two platoon sizes, the method recorded the lowest average per-vehicle fuel consumption of all the control strategies it was benchmarked against.

The motivation is straightforward: despite the growth of electrification, more than one billion gasoline vehicles remain on the world’s roads, and internal combustion engines still dominate personal transport in many countries because of their established supply chains, low costs, and extensive fueling infrastructure. The transportation sector accounts for roughly 24 percent of global carbon dioxide emissions from fuel combustion, according to figures cited by the International Energy Agency in the paper. Vehicle platooning, in which autonomous vehicles travel in close formation using vehicle-to-vehicle communication, exploits the slipstream effect, in which followers sit inside the low-pressure wake of the vehicle ahead. Real-world truck platooning trials, including the European Truck Platooning Challenge and United States Department of Energy initiatives, have shown fuel consumption reductions of up to 10 to 20 percent, and the authors cite potential savings of 5 to 15 percent for passenger vehicles in operating conditions.

What makes gasoline vehicles harder to manage than their electric counterparts is the shape of the engine’s efficiency landscape. A gasoline engine’s brake specific fuel consumption, the amount of fuel burned per unit of work, varies non-linearly with engine speed, torque, and thermal state, and the engine’s sweet spot shifts during acceleration, deceleration, and merging. Electric vehicles can recover energy through regenerative braking; gasoline vehicles cannot, so every application of the friction brakes discards energy that was already spent. The authors argue that most existing platooning research has focused on electric or hybrid drivetrains, leaving gasoline platoons to rely on complementary tricks such as coordinated fuel-cut coasting, optimized mechanical braking, and throttle control that keeps the engine inside its most efficient map region.

The new framework stacks three components. At the core sits an Adaptive Neuro-Fuzzy Inference System, whose 41 parameters are tuned offline by Teaching-Learning-Based Optimization, wrapped inside a model predictive control loop with a ten-step, ten-second horizon. The controller solves a finite-horizon optimization at every one-second step, weighing fuel use, tracking error, throttle smoothness, and estimated carbon monoxide and nitrogen oxide emissions, using a sequential least squares quadratic programming solver warm-started from the previous solution. Above this sits a Quantum-inspired Long Short-Term Memory network, trained only on the leader’s data, which produces a mean-predicted state vector covering velocity, acceleration, and power demand over the next ten seconds. That forecast is broadcast simultaneously to all followers, rather than passed sequentially down the line as in conventional predecessor-following topologies, eliminating propagation delay.

The plug-and-play claim rests on a deliberately thin supervisory layer, implemented in fewer than fifteen lines of Python code, that injects only three new signals into each follower: the leader’s predicted state vector, the instantaneous gap and position index in the chain, and a single binary fuel-cut activation bit triggered when the leader’s predicted acceleration drops below minus 0.2 meters per second squared. A cooperative power-sharing mechanism blends each follower’s local power demand with the leader’s forecasted demand in equal measure, and a predictive gap policy updates the desired spacing, never below a fifteen-meter floor. The local controller, a fixed function with an unchanged parameter set, simply receives modified reference inputs and generates torque as if driving alone. The authors are explicit about what is new and what is inherited: the underlying cooperative adaptive cruise control and model predictive control formulation comes from prior work, while the joint TLBO-optimized ANFIS-plus-MPC decision layer and the simultaneous QLSTM broadcast are the genuine additions.

Validation was purely computational. The team simulated homogeneous platoons of four and six vehicles, each led by one vehicle and followed by three or five, across four driving cycles: the steady-state Highway Fuel Economy Test cycle, the aggressive US06 highway cycle, and two stringent European cycles, Artemis HW130 and Artemis HW150, with sustained speeds of 130 and 150 kilometers per hour. Every scenario was run twenty times with stochastic wind disturbances drawn from a Weibull distribution and road grades drawn from a Rayleigh distribution, applied identically to all vehicles at each time step. Under the HWFET cycle, the proposed method achieved an average per-vehicle fuel consumption of 6.64 liters per 100 kilometers in the four-vehicle platoon and 6.61 in the six-vehicle platoon, beating four benchmarks: Adaptive Multi-Armed Bandit, pure ANFIS, pure MPC, and pure TLBO. The margin over the strongest competitor, pure TLBO, was modest, 0.174 liters per 100 kilometers in the four-vehicle case and 0.193 in the six-vehicle case, but consistent across all runs.

The pattern held under harder conditions. On the US06 cycle, the proposed method averaged 7.44 liters per 100 kilometers per vehicle in the four-vehicle configuration against 7.55 for pure TLBO, and the leader vehicle consumed 7.624 liters per 100 kilometers, roughly 1.7 to 2.4 percent less than the competing leaders. On the Artemis HW150 cycle with six vehicles, follower five consumed 6.554 liters per 100 kilometers versus 6.578 for pure TLBO. The authors attribute the sustained advantage under transient driving primarily to the cooperative power-sharing mechanism, which lets followers smooth their throttle response ahead of upcoming leader maneuvers rather than reacting after the fact. An ablation study on the US06 and Artemis HW130 cycles confirmed that disabling either the QLSTM prediction or the power-sharing layer individually degraded fuel economy by roughly 1 to 2 percent, with the two mechanisms showing overlapping rather than strictly additive contributions.

The emissions picture is more nuanced than the fuel numbers. Nitrogen oxide reductions from leader to last follower reached 5.74 percent in the four-vehicle platoon and 6.23 percent in the six-vehicle platoon, the highest recorded across all methods, but this gradient conceals a genuine trade-off: the leader’s own nitrogen oxide output was higher than that of competing methods, 14.28 grams per kilometer on the HWFET cycle versus 12.97 to 13.35 for the benchmarks, because tighter velocity tracking pushes the engine toward higher throttle openings and richer air-fuel ratios that raise combustion temperature. The authors stress that these are engine-out, pre-catalyst estimates; the model does not include three-way catalytic converter dynamics, so the figures should not be read as tailpipe emissions. Followers benefit disproportionately from the predictive power management, yielding the best fleet-wide reduction gradient despite the leader’s individually higher output.

Scalability and robustness tests rounded out the evaluation. The six-vehicle platoon ran with zero retuning of any local controller, requiring only two new drag-coefficient entries, 0.67 and 0.64, in an external configuration table, which the authors present as an empirical demonstration of the plug-and-play property. In eighty robustness runs spanning all four cycles, both platoon sizes, and five communication scenarios including a fixed 0.3-second broadcast delay, ten percent packet loss, update jitter, and measurement noise, the worst-case inter-vehicle gap never fell below the fifteen-meter floor, and gap-tracking error stayed below 0.01 meters on the steady HWFET cycle and within roughly 4.1 to 4.9 meters on the more dynamic cycles. The QLSTM itself offered only a limited predictive edge, with ten-second-ahead errors modestly below a conventional LSTM, a linear extrapolation, and a naive persistence baseline on held-out test data.

The authors are candid about the boundaries of the work. No hardware-in-the-loop testing, real vehicle data, or realistic actuator dynamics were involved; validation was confined to highway cycles and homogeneous platoons, so extension to heterogeneous fleets remains future work. The drag coefficients are fixed lookup values by platoon position rather than functions of the real-time gap, crosswind effects on drag were not modeled, engine thermal management under platooning was left out, and the gap policy references the leader directly rather than the immediately preceding vehicle. Within those limits, the study offers a quantified, simulation-only case that a lightweight supervisory layer, one that leaves every vehicle’s controller untouched, can extract a small but repeatable fuel economy gain from gasoline platoons, with the benefit growing slightly as the platoon gets longer.

Subject of Research: Predictive energy management for fuel-efficient platooning of gasoline-powered autonomous vehicles

Article Title: Plug-and-play predictive energy management for fuel-efficient platooning of gasoline-powered autonomous vehicles

Article References: Samieiyan, B., & Awasthi, A. (2026). Plug-and-play predictive energy management for fuel-efficient platooning of gasoline-powered autonomous vehicles. Discover Vehicles, 2(1), Article 49. https://doi.org/10.1007/s44465-026-00052-x

Image Credits: AI Generated

DOI: 10.1007/s44465-026-00052-x

Keywords: vehicle platooning, gasoline engines, energy management, model predictive control, ANFIS, QLSTM, autonomous vehicles, fuel consumption, NOx emissions, cooperative control, simulation study, aerodynamic drag