Model-Assisted Learning for Adaptive Cooperative Perception of Connected Autonomous Vehicles

TL;DR

This work investigates unicast-based cooperative perception (CP) for connected autonomous vehicles in a mixed-traffic setting, proposing an adaptive scheme that toggles CP on/off for predetermined CAV pairs and jointly allocates radio and computing resources to maximize the computing efficiency gain under perception delay constraints. A model-assisted multi-agent reinforcement learning (MARL) framework is developed to learn cooperative decisions, while a model-based resource allocator optimizes CPU frequencies and transmission rates in each time slot. Simulation results demonstrate that adaptive CP with model-assisted MARL significantly improves computing efficiency gains and reduces switching costs compared to baselines, validating the approach under dynamic workloads and intermittent RSU coverage. The proposed method offers a scalable solution for efficient CP in real-world V2X networks, with potential impact on resource-constrained, latency-sensitive CP tasks in autonomous driving.

Abstract

Cooperative perception (CP) is a key technology to facilitate consistent and accurate situational awareness for connected and autonomous vehicles (CAVs). To tackle the network resource inefficiency issue in traditional broadcast-based CP, unicast-based CP has been proposed to associate CAV pairs for cooperative perception via vehicle-to-vehicle transmission. In this paper, we investigate unicast-based CP among CAV pairs. With the consideration of dynamic perception workloads and channel conditions due to vehicle mobility and dynamic radio resource availability, we propose an adaptive cooperative perception scheme for CAV pairs in a mixed-traffic autonomous driving scenario with both CAVs and human-driven vehicles. We aim to determine when to switch between cooperative perception and stand-alone perception for each CAV pair, and allocate communication and computing resources to cooperative CAV pairs for maximizing the computing efficiency gain under perception task delay requirements. A model-assisted multi-agent reinforcement learning (MARL) solution is developed, which integrates MARL for an adaptive CAV cooperation decision and an optimization model for communication and computing resource allocation. Simulation results demonstrate the effectiveness of the proposed scheme in achieving high computing efficiency gain, as compared with benchmark schemes.

Figures (11)

• Figure 1: A mixed-traffic autonomous driving scenario.
• Figure 2: Object classification by using a default DNN model.
• Figure 3: Object classification by using a feature-fusion DNN model.
• Figure 4: Examples of 2D performance regions for cooperative CAV pair $k$ at a different shared workload, where $\left[R_{\mathsf{M}},f_{\mathsf{M}}\right]$, $\left[R^{\mathsf{P}}_k(n),f^{\mathsf{P}}_k(n)\right]$, and $\left[R^{\mathsf{D}}_k(n),f^{\mathsf{D}}_k(n)\right]$ are indicated by blue, pink, and green circles.
• Figure 5: An illustration of the root of function $S\left(f_k, \nu\right)$ for $\nu_1<\nu_2$.

Theorems & Definitions (3)

• Theorem 1
• Lemma 1
• proof