# CAN Coach: Vehicular Control through Human Cyber-Physical Systems

**Authors:** M. Nice, S. Elmadani, R. Bhadani, M. Bunting, J. Sprinkle, D. Work  
**arXiv:** [2104.06264](https://arxiv.org/abs/2104.06264)  
**Date:** 2021-04-08  
**Summary page:** https://www.summarizepaper.com/en/paper/2104.06264v1/

## Key points

- The article presents the development and evaluation of a human-in-the-loop cyber-physical system (HCPS) called CAN Coach.
- CAN Coach aims to improve the longitudinal control of individual vehicles in traffic flow by providing feedback to the human driver using radar data available on the controller area network (CAN).
- The study involves six human subjects driving an instrumented vehicle, comparing their ability to achieve a constant time-gap control policy using only visual perception versus augmenting perception with audible feedback from CAN sensor data.
- Results show that adding CAN-based feedback reduces mean time-gap error by 73% and improves consistency by reducing standard deviation of time-gap error by 53%.
- The study also includes a ghost mode where visual perception is removed, and drivers are coached to track a virtual vehicle on the road instead of a physical one. This mode degrades performance for most drivers but shows that they can match velocity with or without CAN-based feedback, although velocity matching does not regulate vehicle spacing.
- The authors demonstrate the viability of dynamic time-gap control.
- Related works informing the formation of CAN Coach include research on controlling traffic waves via Lagrangian control, vehicular-based variable speed limits, modeling and simulation for designing human-in-the-loop vehicle control systems, driver assistance systems for promoting eco-driving at signalized intersections, and field experiments demonstrating human traffic flow stabilization using automated vehicles.
- Overall this study demonstrates that coaching drivers with CAN data can improve their performance on driving tasks while highlighting that removing human perception from the control loop reduces performance at given objectives.
- Therefore CAN Coach is a true HCPS that aims to enhance human driving abilities rather than replace them with the human driver playing a monitoring role while the system carries out control objectives.

## Summary

This article presents the development and evaluation of a human-in-the-loop cyber-physical system (HCPS) called CAN Coach, which aims to improve the longitudinal control of individual vehicles in traffic flow. The system provides feedback to the human driver using radar data available on the controller area network (CAN), specifically relative speed and position information to objects ahead. The study involves six human subjects driving an instrumented vehicle, comparing their ability to achieve a constant time-gap control policy using only visual perception versus augmenting perception with audible feedback from CAN sensor data. Results show that adding CAN-based feedback reduces mean time-gap error by 73% and improves consistency by reducing standard deviation of time-gap error by 53%. The study also includes a ghost mode where visual perception is removed, and drivers are coached to track a virtual vehicle on the road instead of a physical one. This mode degrades performance for most drivers but shows that they can match velocity with or without CAN-based feedback, although velocity matching does not regulate vehicle spacing. The authors demonstrate the viability of dynamic time-gap control. Related works informing the formation of CAN Coach include research on controlling traffic waves via Lagrangian control, vehicular-based variable speed limits, modeling and simulation for designing human-in-the-loop vehicle control systems, driver assistance systems for promoting eco-driving at signalized intersections, and field experiments demonstrating human traffic flow stabilization using automated vehicles. The use of CAN data as a source for analyzing driving behavior is becoming increasingly recognized in research. For instance, previous studies have used it to analyze lane changing, turning, driver categorization and identify aggressive maneuvers in real time. However, unlike CAN Coach introduced here these analyses are not used to close the loop on the human driver in real time. Overall this study demonstrates that coaching drivers with CAN data can improve their performance on driving tasks while highlighting that removing human perception from the control loop reduces performance at given objectives. Therefore CAN Coach is a true HCPS that aims to enhance human driving abilities rather than replace them with the human driver playing a monitoring role while the system carries out control objectives. The authors plan to extend their work in future studies.