Nano-U: Efficient Terrain Segmentation for Tiny Robot Navigation

2026-05-11

AI Summary

This work addresses the challenge of deploying terrain segmentation models on resource-constrained micro-robots, where existing approaches suffer from excessive computational and memory demands. To overcome this limitation, the authors propose MicroFlow, an efficient binary terrain segmentation framework tailored for microcontrollers. It features Nano-U, an ultra-lightweight neural network with only a few thousand parameters, trained via a quantization-aware distillation (QAD) strategy. The inference engine is implemented in Rust as a compiled, interpreter-free runtime that avoids dynamic memory allocation. Deployed on an ESP32-S3 microcontroller, the system achieves real-time terrain perception with low latency and minimal memory footprint, demonstrating strong performance on both the Botanic Garden benchmark and a newly introduced agricultural dataset, TinyAgri. This approach offers a practical, energy-efficient perception solution for low-cost micro-robotics.

Abstract

Terrain segmentation is a fundamental capability for autonomous mobile robots operating in unstructured outdoor environments. However, state-of-the-art models are incompatible with the memory and compute constraints typical of microcontrollers, limiting scalable deployment in small robotics platforms. To address this gap, we develop a complete framework for robust binary terrain segmentation on a low-cost microcontroller. At the core of our approach we design Nano-U, a highly compact binary segmentation network with a few thousand parameters. To compensate for the network's minimal capacity, we train Nano-U via Quantization-Aware Distillation (QAD), combining knowledge distillation and quantization-aware training. This allows the final quantized model to achieve excellent results on the Botanic Garden dataset and to perform very well on TinyAgri, a custom agricultural field dataset with more challenging scenes. We deploy the quantized Nano-U on a commodity microcontroller by extending MicroFlow, a compiler-based inference engine for TinyML implemented in Rust. By eliminating interpreter overhead and dynamic memory allocation, the quantized model executes on an ESP32-S3 with a minimal memory footprint and low latency. This compiler-based execution demonstrates a viable and energy-efficient solution for perception on low-cost robotic platforms.

Problem

• terrain segmentation
• tiny robots
• microcontroller constraints
• autonomous navigation
• outdoor environments

Innovation

• Nano-U
• Quantization-Aware Distillation
• TinyML
• Terrain Segmentation
• Compiler-based Inference