Unveiling Uncertainty-Aware Autonomous Cooperative Learning Based Planning Strategy

TL;DR

This work addresses uncertainty in autonomous cooperative planning (ACP) for multi-vehicle systems, focusing on perception and V2V communication uncertainties that degrade performance and safety. It proposes DRLACP, a GRU-enhanced Soft Actor-Critic framework that learns deterministic time-varying actions under imperfect state information and enforces collision-free motion via a multi-uncertainty formulation. Key contributions include integrating perception and communication uncertainties into a learning-based cooperative MPC paradigm, designing GRU-SAC with a GRU-based actor and dual critics, and validating performance in CARLA showing superior safety, efficiency, and real-time feasibility against baselines. The approach enables robust, scalable planning for uncertain, multi-vehicle environments with potential real-world impact in intelligent transportation systems.

Abstract

In future intelligent transportation systems, autonomous cooperative planning (ACP), becomes a promising technique to increase the effectiveness and security of multi-vehicle interactions. However, multiple uncertainties cannot be fully addressed for existing ACP strategies, e.g. perception, planning, and communication uncertainties. To address these, a novel deep reinforcement learning-based autonomous cooperative planning (DRLACP) framework is proposed to tackle various uncertainties on cooperative motion planning schemes. Specifically, the soft actor-critic (SAC) with the implementation of gate recurrent units (GRUs) is adopted to learn the deterministic optimal time-varying actions with imperfect state information occurred by planning, communication, and perception uncertainties. In addition, the real-time actions of autonomous vehicles (AVs) are demonstrated via the Car Learning to Act (CARLA) simulation platform. Evaluation results show that the proposed DRLACP learns and performs cooperative planning effectively, which outperforms other baseline methods under different scenarios with imperfect AV state information.

Figures (4)

• Figure 1: System architecture of the proposed DRLACP framework.
• Figure 2: Overall processing flow for agent $k$ at time step $t$.
• Figure 3: Predicted error versus steps via proposed GRU-SAC.
• Figure 4: Evaluation of the proposed DRLACP in CARLA.