acceptodds
Under review as a conference paper at ICLR 2027

Event-Triggered Safe-Aware Platoon Re-Sequencing for SoC Balance of Connected Electric Vehicles

Abstract

The position-dependent aerodynamic benefits in connected electric vehicle (EV) platoons provide an opportunity to improve energy efficiency, but they may also result in uneven state-of-charge (SoC) distribution among vehicles during long-distance driving. This paper proposes an event-triggered safe-aware platoon re-sequencing framework to achieve adaptive energy balance optimization while ensuring safe operation in dynamic traffic environments. The platoon re-sequencing problem is formulated by considering the coupling relationships among vehicle positions, aerodynamic energy-saving effects, battery evolution, and traffic-dependent feasibility constraints. To enable safety-aware decision making, a traffic prediction and classification module based on long short-term memory (LSTM) networks and fuzzy c-means (FCM) clustering is developed to predict future traffic conditions and determine whether platoon re-sequencing can be safely executed. Furthermore, an event-triggered mechanism (ETM) is designed to adaptively regulate the execution of re-sequencing actions according to the real-time SoC imbalance and travel progress of the platoon. Two triggering strategies, including a state-aware self-resetting trigger and an adaptive cooldown-based trigger, are developed to dynamically determine appropriate re-sequencing instants under persistent energy imbalance conditions. The proposed framework separates re-sequencing decision generation from action execution, where a reinforcement learning policy provides candidate platoon orders while the traffic prediction module and ETM jointly determine their safe and adaptive execution. The proposed method enables connected EV platoons to autonomously adjust their formation sequence in response to evolving energy distribution and traffic conditions, thereby improving terminal SoC balance while maintaining safe platoon operation.

Primary area:
reinforcement learning
Keywords:
Deep reinforcement learning, Electric vehicles, Energy management, Event-Triggered mechanism

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_vUaCOpenReview ↗
Rating
1
Confidence
2
Informativeness
2
Presentation
1
Soundness
1
Read review
Paper summary

This paper proposes an event-triggered, safety-aware framework for platoon resequencing to balance SoC across connected electric vehicles. The approach decouples candidate order generation (via a DQN policy over the permutation space) from execution, which is gated jointly by a traffic prediction-and-classification module (LSTM + fuzzy C-means) and one of two event-triggered mechanisms (self-resetting or adaptive cooldown) designed to adapt to SoC imbalance and trip progress. The authors provide Zeno-free analyses for the proposed event triggers and report simulation results on platoons of 4–7 vehicles suggesting improved episode rewards and travel distance under the adaptive cooldown trigger.

Critical strengths
  1. The introduction of two event-triggered mechanisms tailored to persistent SoC imbalance, with Zeno-free arguments, is a thoughtful design that addresses timing and frequency of resequencing actions.
  2. The problem setup, definitions, and the role of each architectural component are explained clearly.
Critical weaknesses
  1. The model is simplistic (fixed per-position aerodynamic savings, identical vehicles, no resequencing cost), potentially overstating gains in realistic scenarios.
  2. Missing key baselines: no comparisons against optimization-based resequencing, simple rotation heuristics, or recent combinatorial or sequencing algorithms with guarantees; limited DRL baselines.
  3. No ablation on the LSTM-FCM safety gate, or sensitivity analyses of ETM parameters.
  4. Several crucial quantitative details are relegated to appendices. The main results section is too brief and omits key measurements.
  5. This work’s safety layer is simple and does not simulate or plan detailed maneuvers, limiting the safety claim and realism.

The main text contains few figures and tables, with most key experimental results relegated to the appendix, while the overall experimental evidence is insufficient to convincingly demonstrate the effectiveness of the proposed method. In addition, the paper devotes excessive space to descriptions that are weakly related or unrelated to its key innovations, at the expense of presenting the core contributions and supporting evidence.

Questions for authors
  1. What baselines were attempted (e.g., periodic/round-robin leader rotation, simple optimization-based min-variance resequencing, bootstrapped/dueling DQN or PPO/TRPO)? Please include quantitative comparisons.
  2. Are resequencing costs (time, control effort, potential energy penalties during lane changes) modeled? If not, how might including them affect the ETM thresholds and policy behavior?
  3. The Zeno-free proof for ETM1 introduces an auxiliary margin η in the appendix. Is this margin implemented in the experiments, and how are κ and other constants chosen in practice?
  4. Are aerodynamic coefficients e_j calibrated from literature and do they depend on inter-vehicle gap and speed? If fixed, what is the impact of varying e_j or heterogeneous vehicles/batteries on results?
Reviewer_YFLmOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper studies state-of-charge imbalance in connected electric vehicle platoons arising from differences in aerodynamic energy savings across platoon positions. It proposes an event-triggered re-sequencing framework that separates candidate-order generation from execution. It combines DQN-based re-sequencing, LSTM-FCM traffic prediction, and two event-triggering mechanisms to regulate vehicle ordering. Experiments compare different ordering policies and triggering strategies in terms of reward, travel distance, and re-sequencing frequency.

Critical strengths

The problem has clear practical relevance. Aerodynamic benefits differ across platoon positions, so maintaining a fixed order can cause some vehicles to reach their minimum SoC earlier than others, limiting the usable travel distance of the entire platoon. The paper studies how dynamic re-sequencing can redistribute the energy consumption burden to improve terminal SoC balance. This objective has a clear practical motivation: making fuller use of the energy available across the platoon.

Critical weaknesses
  1. Limited methodological novelty. The proposed framework mainly integrates existing techniques, including DQN, LSTM-FCM, and event-triggered control, with limited methodological innovation. The experimental baselines are mostly traditional reinforcement learning methods and earlier approaches, with the most recent baseline dating back to 2022. More recent task-specific methods [1, 2] are not included, making it difficult to establish the novelty and competitiveness of the proposed framework.
  2. Insufficient experimental validation. The experiments lack ablations of the proposed event-triggering and traffic prediction mechanisms, as well as comparisons with simpler execution strategies such as periodic triggering or fixed cooldown. Without comparisons against DQN without these modules, it remains unclear whether the proposed framework provides additional benefits beyond the underlying DQN policy. The individual contributions of the core components are not adequately validated.

[1] Congestion-aware platoon re-sequencing optimization for electric vehicles using deep reinforcement learning

[2] Dynamic Platoon Re-Sequencing for Electric Vehicles Based on Bootstrapped DQN

Questions for authors
  1. What are the main methodological contributions of the proposed framework beyond integrating existing DQN, LSTM-FCM, and event-triggered control techniques?
  2. Can the authors provide comparisons with more recent task-specific methods?
  3. Can the authors provide ablation studies to evaluate the individual contributions of the event-triggering and traffic prediction modules? In particular, how does the proposed framework compare with standard DQN using periodic re-sequencing or fixed cooldown strategies under the same experimental settings?
Reviewer_HqRzOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
1
Soundness
1
Read review
Paper summary

This paper studies event-triggered re-sequencing for electric vehicle platoons. It combines a DQN policy, traffic prediction, and two event-triggering mechanisms to select vehicle orders and regulate their execution for SoC balancing.

Critical strengths

The paper addresses a practically relevant problem concerning unequal energy consumption in electric vehicle platoons.

Critical weaknesses
  1. The overall layout of the article is extremely unsatisfactory! The experimental section is only introduced on the final page. There are no diagrams or tables in the main text to explain the methods and demonstrate the effectiveness.

  2. Equations 17–18 divide by the previous SoC variance, but all vehicles start with identical SoC, giving zero initial variance; the paper does not clearly specify how this case is handled.

  3. The reported travel distances appear inconsistent with the simulation parameters. With an 85 kWh battery, initial SoC of 1, minimum SoC of 0, and the stated energy model, even a vehicle continuously in the leading position should travel approximately 421 km before depletion, whereas the tables report approximately 267–284 km.

  4. DQN does not consistently outperform the compared policies, and its action space contains vehicle permutations, making the scalability claim difficult to support with experiments limited to four to seven vehicles.

Questions for authors
  1. How is the reward computed when the SoC variance is zero or close to zero, especially at the beginning of each episode?

  2. Can the authors reconcile the reported travel distances with the stated battery capacity, energy-consumption model, initial SoC, and termination condition?

  3. What are the terminal SoC standard deviation and maneuver-level safety outcomes, including comparisons with fixed-order and fixed-interval strategies?

  4. How do DQN’s training time, inference time, and performance scale with platoon size given the factorial action space, and why is DQN preferable to the simpler policies in the reported results?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.