Low-cost miniature vehicle uses split hardware for safer neural driving
A Hardware-Partitioned micro-ROS Mini-Vehicle for Operator-Supervised Neural Driving
Robotics
Summary
Controlling small self-driving cars with neural networks can be tricky because safety and real-time control need to be very fast and reliable. The authors created a little car with split hardware parts: one node with a camera sends video wirelessly to a more powerful computer, which does the brain-like decision making, while a separate controller on the car handles fast and safe driving commands. This setup keeps safety-critical tasks on the car itself, reducing risks from delays in wireless communication. Their tests showed the system works well, with no crashes in multiple runs and improved accuracy in tracking movement using sensor fusion.
What this means in practice
- •For robotics hardware developers: Build and test embedded autonomous vehicles that run neural networks off-board while maintaining safety-critical control onboard for real-time actuation.
- •For home automation engineers: Integrate low-cost wireless sensing and real-time control architectures from mini vehicles into remote robotic devices requiring both safety and AI-driven decision-making.
Authors
Gustavo Claudio Karl Couto, Eric Aislan Antonelo, Gabriel George Zipperer
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.