A Brain-Inspired Perception-Decision Driving Model Based on Neural Pathway Anatomical Alignment
Haidong Wang, Pengfei Xiao, Ao Liu, Qia Shan, Jianhua Zhang · 2025
Read A Brain-Inspired Perception-Decision Driving Model Based on Neural Pathway Anatomical Alignment by Haidong Wang, Pengfei Xiao, Ao Liu, Qia Shan, Jianhua Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-16027.txt.
Introduction
Haidong Wang and co-authors design a self-driving model shaped by the brain, in a 2025 paper. Most driving stacks lean on sensors plus hand-written rules. Those stacks can be hard to read and brittle in messy traffic. The brain, by contrast, links sight to choice through layered pathways that stay robust.
Their framework, called BID, copies that layout. A brain-inspired sensing stage reads the road scene with high efficiency. A second stage, also modeled on brain pathways, turns what is seen into driving choices. The stages line up with the paths the brain itself uses to move signals along. In tests across driving tasks, the model beat prior methods and drove end to end on its own. The authors say the design adds clarity and strength to self-driving tech.
Catalog record
| CREATOR | Haidong Wang, Pengfei Xiao, Ao Liu, Qia Shan, Jianhua Zhang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-16027
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Brain-Inspired Perception-Decision Driving Model Based on Neural Pathway Anatomical Alignment Haidong Wang, Pengfei Xiao, Ao Liu, Qia Shan, Jianhua Zhang Abstract In the realm of autonomous driving, conventional approaches for vehicle perception and decision-making primarily rely on sensor input and rule-based algorithms. However, these methodologies often suffer from lack of interpretability and robustness, particularly in intricate traffic scenarios. To tackle this challenge, we propose a novel brain-inspired driving (BID) framework. Diverging from traditional methods, our approach harnesses brain-inspired perception technology to achieve more efficient and robust environmental perception. Additionally, it employs brain-inspired decision-making techniques to facilitate intelligent decision-making. The experimental results show that the performance has been significantly improved across various autonomous driving tasks and achieved the end-to-end autopilot successfully. This contribution not only advances interpretability and robustness but also offers fancy insights and methodologies for further advancing autonomous driving technology. I Introduction Autonomous driving [ 1 , 2 ] is an advanced technology that intelligent vehicles perceive road environments through onboard sensor systems, autonomously plan driving routes, and control vehicles to reach predetermined destinations. Its technical system generally includes three major parts: environmental perception, decision planning, and vehicle control [ 3 ] , involving multiple research fields such as computer science, mathematics, mechanical engineering, control science, and psychology [ 4 ] . However, the current autonomous driving systems still suffer from insufficient interpretability due to the existence of “black box” nature of deep learning models [ 5 ] , greatly limiting the credibility and widespread application of various perception and decision-making methods in practical engineering. Even though the use of generative adversarial networks [ 6 ] to generate explanatory data related to decision-making has been attempted, the quality of such data is often substandard, and the training process is quite challenging. Furthermore, modern urban traffic conditions are characterized by high dynamics and strong uncertainty. When the environment is complex, the autonomous driving system may experience performance degradation. Therefore, to overcome the limitations of existing methods, there is a need to design more interpretable and robust autonomous driving systems, thereby ensuring the safety of vehicle operation. Fig. 1: Perception-decision network diagram based on neural pathway anatomical alignment In this work, we propose a novel brain-inspired driving (BID) framework inspired by brain perception and decision-making. Unlike previous methods, our proposed BID expert does not rely on manually formulated rules and demonstrates strong interpretability and robustness. Our work involves constructing a brain-inspired perception network for extracting environmental features and a brain-inspired decision-making network for generating driving decisions for the target vehicle, in order to train the BID expert [ 7 ] . Its structure is as shown in Fig. 1 . Specifically, the brain-inspired perception network consists of a visual recognition [ 8 ] , a motion perception, and a multimodal fusion [ 9 ] . The brain-inspired decision-making network comprises a decision generation [ 10 ] and a decision evaluation. Driving decisions encompass multiple driving actions of the target vehicle, and the BID agent directly imitates expert behavior. Overall, our main contribution lies in proposing a novel brain-inspired perception model and a brain-inspired decision-making model for autonomous driving, aiming to achieve a more robust and interpretable autonomous driving system. II Related Work II-A Brain-Inspired Perception The human visual cortex possesses remarkable environmental perception capabilities, serving as a crucial component of the central nervous system and responsible for transforming visual signals into comprehensible information [ 11 ] . Traditional visual encoding models are limited by the use of either handcrafted or deep learning features merely, despite their individual advantages [ 12 ] . To integrate the two, Cui et al. [ 8 ] proposed the GaborNet-VE model, which combines Gabor features with deep learning to form an efficient visual encoding framework. However, this model requires significant parameter adjustments across different tasks, and the interpretability of its deep learning component remains inadequate [ 9 ] . In contrast, our proposed BID model demonstrates outstanding performance in revealing decision-making mechanisms and exhibiting strong generalization capabilities. II-B Brain-Inspired Decision In practical environments, agents often need to handle continuous state spaces, while traditional reinforcement learning models, such as Q learning methods, are more fit to discrete states [ 13 ] . To address this issue, Zhao et al. [ 14 ] proposed the prefrontal cortex and basal ganglia method, which subdivides continuous states and utilizes a continuous function to capture temporal reward. However, the design of continuous reward functions remains a challenge. In non-Gaussian and nonlinear environments, traditional methods struggle to cope with their nonlinear and non-Gaussian characteristics [ 15 ] . Inspired by human decision-making in brain, some researchers [ 10 ] designed an autonomous computation layer accroding to a cognitive dynamic system (CDS), providing agents in NGNLE with stronger decision-making capabilities. Although these methods attempt to mimic the decision-making process of the human brain, structural differences lead to a lack of transparency in their processes. Therefore, it is crucial to construct a network architecture that align with the anatomical structure of human brain. III Method III-A Task Formulation We use imitation learning (IL) to train the proposed BID. In IL, expert first demonstrates a driving action a_{t} (such as steering, brake and throttle) in response to an input frame f_{t} based on the expert’s policy \pi(f_{t}) . It reflects the expert’s driving skills, judgement, and behavior. The core idea of IL is to train the proposed BID agent to replicate the expert’s actions by learning from these observed interactions. Fig. 2: The network architecture of the BID agent is primarily composed of the perception network and the decision network. The perception module includes both the dorsal and ventral pathways, which allow the system to understand and process its surroundings. The decision network consists of five modules that represent different regions of the prefrontal cortex: the medial prefrontal cortex (MPC), orbital prefrontal cortex (OPC), caudal prefrontal cortex (CPC), dorsal prefrontal cortex (DPC), and ventral prefrontal cortex (VPC). Specifically, the input to the system is an RGB image f_{t} , which represents the current visual scene. The output is the action A_{t} , which controls the ego-vehicle’s maneuvering directly. The action A_{t} includes the steering angle A_{t}^{s} and acceleration A_{t}^{a} of the vehicle, making BID a brain-inspired and end to end autonomous driving model. Besides, in the construction process of training dataset, the driving actions and input frames are paired. Before training the BID using IL, we first collect a series of frame and action pairs (f_{t},a_{t}) produced by expert demonstrations. These data are then utilized to train the BID policy \pi_{\theta}(f_{t}) , which approximates the expert’s policy. In the validating phase, we no longer utilize the expert’s strategy, but only the trained BID policy. III-B Neural Aligned BID Inspired by the brain’s remarkable capabilities in perception and decision-making, we have proposed the perception and decision network based on neural pathway anatomical alignment to simulate the information processing mechanisms of brain in an autonomous driving system. This approach is designed to significantly improve both the interpretability and robustness of the system. III-B 1 Perception Network As shown in Fig. 2 , our brain-inspired perception network begins by capturing precise input data from the surrounding environment, which includes essential traffic elements such as roads, vehicles, pedestrians,