A Hardware-Partitioned micro-ROS Mini-Vehicle for Operator-Supervised Neural Driving

Gustavo Claudio Karl Couto; Eric Aislan Antonelo; Gabriel George Zipperer

cs.ROOct 11, 2026

A Hardware-Partitioned micro-ROS Mini-Vehicle for Operator-Supervised Neural Driving

Authors: Gustavo Claudio Karl Couto, Eric Aislan Antonelo, Gabriel George Zipperer

Organizations: Automation and Systems Engineering Department Federal University of Santa Catarina, Florianópolis, Brazil

Abstract

We present a low-cost miniature Ackermann vehicle for embedded neural-driving research, organized as a hardware-partitioned perceive-decide-act agent: an ESP32 camera node streams frames over Wi-Fi to a host computer, where neural inference and state estimation run, while an ESP32-S3 chassis controller executes safety-critical actuation in a 100 Hz per-wheel proportional-integral-derivative (PID) loop beneath a micro-ROS client. The partition isolates real-time control and safety on the vehicle while learning-based components run off-board, supervised by an operator from a gamepad. We characterize the complete pipeline during real closed-loop neural driving. Wireless sensing, rather than neural inference, dominates latency, with a 59.5 ms median capture-to-actuation delay; the platform absorbed wireless variability by holding the last safe command under a firmware watchdog, operating across twelve autonomous sessions with zero collisions or manual interventions; and encoder-inertial-measurement-unit (IMU) fusion cut odometry loop-closure error by 22%. These results indicate that carefully partitioned microcontroller-class vehicles support reproducible experiments in embedded AI, wireless control, and operator-supervised autonomous driving. The platform software is open source.

Figures & tables

Fig. 1: The vehicle: an off-the-shelf Ackermann chassis carrying two microcontrollers—an ESP32-S3 chassis controller and an ESP32 camera module on a short mast. There is no on-board single-board computer.

Fig. 2: Hardware/software architecture: the ESP32-S3 chassis controller, the ESP32 camera node, the Wi-Fi link, the host-side estimator, policy, and operator interface. Solid edges: wired links; dashed: across Wi-Fi; thick orange: the

100\,\mathrm{Hz}

control loop.

Component	approx. US$
Ackermann aluminum chassis (incl. $2\times$ MD520
encoder motors and steering servo) †	$108$
micro-ROS control board (ESP32-S3, IMU)	$64$
ESP32 Wi-Fi camera module (OV2640)	$15$
LiPo battery 3S $11.1\,\mathrm{V}$ $2200\,\mathrm{mAh}$ †	$47$
Vehicle total	$\mathbf{\approx 234}$

TABLE I: Approximate bill of materials for the miniature vehicle platform.

Fig. 3: Functional perceive–decide–act view of the platform, one loop per module: camera streaming, host-side policy execution, chassis control, and operator supervision closing the loop.

Fig. 4: The printed $3.2{\times}2.8\,\mathrm{m}$ track and the four evaluation routes (color), with the trajectories driven by the deployed neural policy (dark; mean route error $3.9$ – $6.5\,\mathrm{cm}$ ). The white dot marks the shared start pose.

Stage	Min	Median	p95	Max
Camera link (capture $\rightarrow$ host)	$26.9$	$27.4$	$30.6$	$44.1$
JPEG decode	$0.6$	$0.8$	$1.1$	$2.2$
Host queueing (decode $\rightarrow$ inference)	$0.5$	$13.0$	$26.3$	$74.8$
Policy inference	$2.0$	$2.5$	$3.9$	$5.2$
Command transport (host $\rightarrow$ chassis)	$1.3$	$3.4$	$65.6$	$234.6$
Chassis (command $\rightarrow$ motor PWM)	$1.0$	$7.4$	$9.6$	$45.0$

TABLE II: Capture-to-actuation latency measured during closed-loop neural-control operation (

528

frames; all values in ms).

Fig. 5: Mean cross-track error in the digital twin as a function of injected sensing delay, for the deployed policy on the four routes. The dotted line marks the measured median camera-link delay.

Fig. 6: Normalized speed-loop step response at autonomous-session starts. Thin lines: individual steps; thick line: their mean; dashed: the commanded step; band: 10–90th percentile.

Finding	Evidence	Design implication
Wireless link dominates latency	camera link $27\,\mathrm{ms}$ vs. $2.5\,\mathrm{ms}$ inference	optimize transport before the network
End-to-end timing has heavy tails	p95 twice the median; gaps up to $1.04\,\mathrm{s}$	timestamped processing; watchdog protection
Heading dominates odometry drift	EKF fusion cuts loop-closure error $22\%$	gyroscope fusion pays even at this scale
Encoder quantization limits low-speed PID	sustained $\pm 20\%$ speed ripple	higher-resolution encoders for lower speeds

TABLE III: Design lessons from the characterization.

Robotics

Robotics studies the design and control of machines that perceive, move, and act in physical environments.

See all 755 papers

A Low-Cost, Open Platform for End-to-End Autonomous Driving on a Miniature Ackermann Vehicle
Sep 3, 2026Gustavo Claudio Karl Couto, Eric Aislan Antonelo, Gabriel George ZippererEnd-to-End Autonomous Driving Sim-to-Real Transfer
Department of Automation and Systems Engineering Federal University of Santa Catarina (UFSC) – Florianópolis – SC – Brazil
TinyNav: End-to-End TinyML for Real-Time Autonomous Navigation on Microcontrollers
Mar 10, 2026Pooria Roy, Nourhan Jadallah. Tomer Lapid, Shahzaib Ahmad +2Robot Navigation On-Device Inference
Queen’s University
Trajectory Planning and Control near the Limits: an Open Experimental Benchmark on the RoboRacer Platform
May 19, 2026Mattia Piccinini, Patrick Zambiasi, Aniello Mungiello +3Autonomous Driving Benchmarks Robot Trajectory Tracking
Professorship of Autonomous Vehicle Systems, Technical University of Munich, 85748 Garching, Germany; Munich Institute of Robotics and Machine Intelligence (MIRMI). · Avilus GmbH, Germany. · Department of Information Technology and Electrical Engineering (DIETI), University of Naples Federico II, Naples 80125, Italy. +1

Sep 3, 2026J/K move · Enter open · S save

A Low-Cost, Open Platform for End-to-End Autonomous Driving on a Miniature Ackermann Vehicle

Gustavo Claudio Karl Couto, Eric Aislan Antonelo, Gabriel George Zipperer

Department of Automation and Systems Engineering Federal University of Santa Catarina (UFSC) – Florianópolis – SC – Brazil

End-to-End Autonomous Driving Sim-to-Real Transfer Imitation Learning Behavior Cloning Autonomous Driving Simulation

Open paper AskPDF arXiv

This paper presents a low-cost, open experimental platform for research in end-to-end autonomous driving with miniature Ackermann vehicles. The platform combines a physical vehicle, a printed urban track, data collection tools, trajectory registration, and a Webots digital twin, enabling controlled experiments that connect simulation-based autonomous-driving methods to real-world execution. As a first baseline, we implement command-conditioned behavior cloning, in which a neural policy receives an on-board camera image and a high-level navigation command and outputs steering and speed. The system is evaluated both on the physical vehicle and in simulation. In real closed-loop experiments, the learned policy follows lanes and executes commanded turns, reaching a mean cross-track error of 6.1 cm with respect to the reference route, close to the 4.7 cm observed in human demonstrations. In the digital twin, camera field of view has a strong effect on performance, reducing the mean cross-track error from 35.6 to 3.3 cm when widened from 58 to 120 degrees. Using the digital twin to generate synthetic driving data and a learned sim-to-real image translator to reduce the appearance gap, we further show that a higher-capacity policy trained on this synthetic data combined with real demonstrations is the only configuration that completes all four track routes in closed loop, whereas the compact baseline and the same network trained on real data alone complete fewer. These results establish the open platform as a practical testbed for sim-to-real studies and provide an initial command-conditioned imitation-learning baseline; we release it to support reproducible research.

Abstract

Robotics

Explore similar work

A Low-Cost, Open Platform for End-to-End Autonomous Driving on a Miniature Ackermann Vehicle

TinyNav: End-to-End TinyML for Real-Time Autonomous Navigation on Microcontrollers

Trajectory Planning and Control near the Limits: an Open Experimental Benchmark on the RoboRacer Platform