terrain-classification

😎 A curated list of papers, tools, and libraries for traversability analysis and terrain classification, segmentation in robotic navigation

Terrain-aware locomotion pipeline for quadruped robots - 95% navigation success, 50% fall reduction using depth camera perception, ML terrain classification, and adaptive footstep planning with MoveIt2

An SVM-based image classifier that identifies terrain types (plains, highlands, water) from topographic maps using distinct color patterns.

This method uses unsupervised and soft self-supervision for auditory Martian terrain classification , based on data from the Perseverance Rover. Next Up, we plan to make this system completely self-supervised with martian terrain images

Field assessment of force-torque sensors for planetary rover navigation, including terrain classification from force-torque vs IMU data. Code for Gerdes et al., JIRS 2025, using BASEPROD.

Deep learning-based terrain classification on the DeepSat-6 dataset using Custom CNN and transfer learning (VGG16, ResNet50). Achieved up to 97.5% accuracy for automated land-cover classification.

Adaptive multimodal locomotion on AgileX Limo Pro: real-time terrain classification and automatic mode switching.

CNN-based terrain traversability classifier for autonomous rovers using RUGD dataset

Multi-task perception network for quadruped robots: traversability segmentation, terrain classification and anchor-free object detection in one PyTorch forward pass, exported to ONNX/TensorRT for Jetson Orin.

MANTLE: Multi-task Adaptive Network for Terrain and Landform Extraction, dual-modality planetary perception via a shared frozen DINOv2 backbone with modular uplink-compatible task-specific heads.

DTMTC: Domain Transfer for Martian Traversability Classification

SAR autofocus and terrain classification using ResU-Net and CNN - JARS 2024 (PhD/MSc research by Mohamed Sakr)

Terrain Recognition — CNN-based terrain classification (grass, sand, rocks, etc.) for SIH 2024 hackathon.

Vision-based terrain classification for quadruped robots using ResNet18 transfer learning and sim-to-real transfer via domain randomization in MuJoCo.

Classfication in terrain images using transfer learning

ESP32 rover that detects surface type from IMU vibration and adapts speed/torque in real time to prevent slipping.