Reinforcement Learning Enhancement Using Vector Semantic Representation and Symbolic Reasoning for Human-Centered Autonomous Emergency Braking

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Asodia, Vinal, Sharifi, Iman, Fallah, Saber
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911422329389056
author Asodia, Vinal
Sharifi, Iman
Fallah, Saber
author_facet Asodia, Vinal
Sharifi, Iman
Fallah, Saber
contents The problem with existing camera-based Deep Reinforcement Learning approaches is twofold: they rarely integrate high-level scene context into the feature representation, and they rely on rigid, fixed reward functions. To address these challenges, this paper proposes a novel pipeline that produces a neuro-symbolic feature representation that encompasses semantic, spatial, and shape information, as well as spatially boosted features of dynamic entities in the scene, with an emphasis on safety-critical road users. It also proposes a Soft First-Order Logic (SFOL) reward function that balances human values via a symbolic reasoning module. Here, semantic and spatial predicates are extracted from segmentation maps and applied to linguistic rules to obtain reward weights. Quantitative experiments in the CARLA simulation environment show that the proposed neuro-symbolic representation and SFOL reward function improved policy robustness and safety-related performance metrics compared to baseline representations and reward formulations across varying traffic densities and occlusion levels. The findings demonstrate that integrating holistic representations and soft reasoning into Reinforcement Learning can support more context-aware and value-aligned decision-making for autonomous driving.
format Preprint
id arxiv_https___arxiv_org_abs_2602_05079
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Reinforcement Learning Enhancement Using Vector Semantic Representation and Symbolic Reasoning for Human-Centered Autonomous Emergency Braking
Asodia, Vinal
Sharifi, Iman
Fallah, Saber
Robotics
Machine Learning
The problem with existing camera-based Deep Reinforcement Learning approaches is twofold: they rarely integrate high-level scene context into the feature representation, and they rely on rigid, fixed reward functions. To address these challenges, this paper proposes a novel pipeline that produces a neuro-symbolic feature representation that encompasses semantic, spatial, and shape information, as well as spatially boosted features of dynamic entities in the scene, with an emphasis on safety-critical road users. It also proposes a Soft First-Order Logic (SFOL) reward function that balances human values via a symbolic reasoning module. Here, semantic and spatial predicates are extracted from segmentation maps and applied to linguistic rules to obtain reward weights. Quantitative experiments in the CARLA simulation environment show that the proposed neuro-symbolic representation and SFOL reward function improved policy robustness and safety-related performance metrics compared to baseline representations and reward formulations across varying traffic densities and occlusion levels. The findings demonstrate that integrating holistic representations and soft reasoning into Reinforcement Learning can support more context-aware and value-aligned decision-making for autonomous driving.
title Reinforcement Learning Enhancement Using Vector Semantic Representation and Symbolic Reasoning for Human-Centered Autonomous Emergency Braking
topic Robotics
Machine Learning
url https://arxiv.org/abs/2602.05079