Final-Command Constraint Enforcement for Hybrid Learning-Predictive Vehicle Lateral Control: An Offline Audit on Real Driving Logs with Real-Time Feasibility Benchmarking

Abstract Hybrid vehicle controllers that combine learned and predictive steering proposals require constraint enforcement at the final actuator command. This study presents an offline constraint-audit framework for a lateral-acceleration inner loop integrating implicit Q-learning (IQL), model predictive control (MPC), and final high-order control barrier function (HOCBF) projection. A platform-specific affine steering-response model and calibration-derived residual margin define the admissible torque interval, onto which the fused IQL–MPC command is projected. Experiments use three public vehicle archives from the commaSteeringControl dataset, comprising 899,188 observations, 756,941 eligible transitions, and 30,892 held-out states for command auditing. The response model achieved held-out acceleration root-mean-square errors of 0.02715, 0.02819, and 0.02075 m/s² for Genesis, Kia, and Toyota, respectively, with residual-envelope coverage near the nominal 95% level. Under the default configuration, final-filter projection yielded zero model-inequality violations at the audited states across all three platforms, compared with pre-filter rates of 5.20%, 7.10%, and 0.08%. Ablation and sensitivity analyses characterized the contributions of IQL, MPC, and residual calibration, while alternative regression specifications revealed instability in the fitted torque-response coefficient. Sequential testing on a desktop-class central processing unit (CPU) measured a mean pipeline latency of 3.097 ms, with no deadline misses at 100 Hz; HOCBF projection accounted for less than 1% of mean computational time. The results indicate that applying HOCBF projection after command fusion improves model-conditional constraint admissibility with limited computational overhead in the evaluated offline setting.

Authors

Institutions

Publication Details

Journal
Engineering Research Express
Published
2026-11-17
DOI
https://doi.org/10.1088/2631-8695/aec355
Primary Topic
Vehicle Dynamics and Control Systems
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Final-Command Constraint Enforcement for Hybrid Learning-Predictive Vehicle Lateral Control: An Offline Audit on Real Driving Logs with Real-Time Feasibility Benchmarking

Jian Zhang
2 links
Engineering Research Express
Vehicle Dynamics and Control Systems
article

Final-Command Constraint Enforcement for Hybrid Learning-Predictive Vehicle Lateral Control: An Offline Audit on Real Driving Logs with Real-Time Feasibility Benchmarking

Jian Zhang
article en

Abstract

Abstract Hybrid vehicle controllers that combine learned and predictive steering proposals require constraint enforcement at the final actuator command. This study presents an offline constraint-audit framework for a lateral-acceleration inner loop integrating implicit Q-learning (IQL), model predictive control (MPC), and final high-order control barrier function (HOCBF) projection. A platform-specific affine steering-response model and calibration-derived residual margin define the admissible torque interval, onto which the fused IQL–MPC command is projected. Experiments use three public vehicle archives from the commaSteeringControl dataset, comprising 899,188 observations, 756,941 eligible transitions, and 30,892 held-out states for command auditing. The response model achieved held-out acceleration root-mean-square errors of 0.02715, 0.02819, and 0.02075 m/s² for Genesis, Kia, and Toyota, respectively, with residual-envelope coverage near the nominal 95% level. Under the default configuration, final-filter projection yielded zero model-inequality violations at the audited states across all three platforms, compared with pre-filter rates of 5.20%, 7.10%, and 0.08%. Ablation and sensitivity analyses characterized the contributions of IQL, MPC, and residual calibration, while alternative regression specifications revealed instability in the fitted torque-response coefficient. Sequential testing on a desktop-class central processing unit (CPU) measured a mean pipeline latency of 3.097 ms, with no deadline misses at 100 Hz; HOCBF projection accounted for less than 1% of mean computational time. The results indicate that applying HOCBF projection after command fusion improves model-conditional constraint admissibility with limited computational overhead in the evaluated offline setting.

Engineering Research Express
Anhui Wenda University of Information Engineering (CN)
Vehicle Dynamics and Control Systems
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon