This software package contains the Model04 EIGT-Net implementation developed for the MPGSD-AEB study. The package includes the trained model parameters, PyTorch model implementation, graph definition, MATLAB/Simulink forward-inference model, and fixed-step data feeder for hardware-in-the-loop preparation. The model processes a 17-frame history of seven traffic nodes. Each node contains four motion features, and each directed edge contains two relative-motion features. The graph uses twelve directed edges: the three far-field vehicles connect to the three near-field vehicles, and the three near-field vehicles connect to the ego vehicle. Two 64-unit graph layers with relation-aware edge gates are followed by a standard 64-unit LSTM and prediction heads for average deceleration, maximum deceleration, and scene probabilities. The Simulink implementation uses a 5 ms fixed simulation step and updates the neural-network history at 120 ms keyframe intervals after the complete history window is available. The package also includes the predictive safety-distance AEB logic and outputs for model-based braking, TTC comparison, warnings, predicted deceleration, scene probabilities, safety margin, and target selection. Raw HighD data, complete training and validation manifests, TruckSim binaries, TSMaster configuration files, and license-dependent runtime environments are not included. The corresponding version-controlled source repository is available at: https://github.com/SihanChenstudy/MPGSD-AEB Release version: v1.0.0.
This software package contains the Model04 EIGT-Net implementation developed for the MPGSD-AEB study. The package includes the trained model parameters, PyTorch model implementation, graph definition, MATLAB/Simulink forward-inference model, and fixed-step data feeder for hardware-in-the-loop preparation. The model processes a 17-frame history of seven traffic nodes. Each node contains four motion features, and each directed edge contains two relative-motion features. The graph uses twelve directed edges: the three far-field vehicles connect to the three near-field vehicles, and the three near-field vehicles connect to the ego vehicle. Two 64-unit graph layers with relation-aware edge gates are followed by a standard 64-unit LSTM and prediction heads for average deceleration, maximum deceleration, and scene probabilities. The Simulink implementation uses a 5 ms fixed simulation step and updates the neural-network history at 120 ms keyframe intervals after the complete history window is available. The package also includes the predictive safety-distance AEB logic and outputs for model-based braking, TTC comparison, warnings, predicted deceleration, scene probabilities, safety margin, and target selection. Raw HighD data, complete training and validation manifests, TruckSim binaries, TSMaster configuration files, and license-dependent runtime environments are not included. The corresponding version-controlled source repository is available at: https://github.com/SihanChenstudy/MPGSD-AEB Release version: v1.0.0.
Sihan Chen· Figshare· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
