Energy-Efficient Autonomous Driving with Adaptive Perception and Robust Decision

TL;DR

EneAD addresses the energy bottleneck in autonomous driving by optimizing perception energy through an adaptive perception module and ensuring robust decisions through a regularized reinforcement learning framework. It employs a task-specific scenario-difficulty classifier and a transferable, Bayesian-optimized knob-tuning process to select low-computation configurations without compromising accuracy. The decision module models driving as an MDP and uses a regularized, parameterized-action RL approach to maintain stability under perception perturbations. Experimental results on real and synthetic data show substantial perception-energy reductions (1.9x–3.5x) and driving-range gains (3.9%–8.5%), while preserving safety and improving comfort in diverse traffic scenarios.

Abstract

Autonomous driving is an emerging technology that is expected to bring significant social, economic, and environmental benefits. However, these benefits come with rising energy consumption by computation engines, limiting the driving range of vehicles, especially electric ones. Perception computing is typically the most power-intensive component, as it relies on largescale deep learning models to extract environmental features. Recently, numerous studies have employed model compression techniques, such as sparsification, quantization, and distillation, to reduce computational consumption. However, these methods often result in either a substantial model size or a significant drop in perception accuracy compared to high-computation models. To address these challenges, we propose an energy-efficient autonomous driving framework, called EneAD. In the adaptive perception module, a perception optimization strategy is designed from the perspective of data management and tuning. Firstly, we manage multiple perception models with different computational consumption and adjust the execution framerate dynamically. Then, we define them as knobs and design a transferable tuning method based on Bayesian optimization to identify promising knob values that achieve low computation while maintaining desired accuracy. To adaptively switch the knob values in various traffic scenarios, a lightweight classification model is proposed to distinguish the perception difficulty in different scenarios. In the robust decision module, we propose a decision model based on reinforcement learning and design a regularization term to enhance driving stability in the face of perturbed perception results. Extensive experiments evidence the superiority of our framework in both energy consumption and driving performance. EneAD can reduce perception consumption by 1.9x to 3.5x and thus improve driving range by 3.9% to 8.5%

Figures (14)

• Figure 1: Knob Setting in Perception System
• Figure 2: Framework Overview
• Figure 3: Model Architecture of Our Scenario Classification Model, which includes a Swin-T network to encode image data and a classification head with dropout layers to calculate a perception difficulty level $\hat{y}$ ($\hat{y} \in {1,2,\dots,k}$) of each image data $x$. In the inference phase, the classification head is run $T$ times to obtain an uncertainty value $u$. If $u$ is larger than a threshold $G$, the difficulty level will be set to $k$ (i.e., the highest difficulty level).
• Figure 4: Model Architecture of Our Reinforcement Learning Model. The actor network outputs three action values corresponding to three lane-changing decisions, and the value network outputs the Q values of the three actions. The action with the highest Q value is the final action performed by the autonomous vehicle.
• Figure 5: Evaluation of Perception Consumption