Autonomous Forest Inventory with Legged Robots: System Design and Field Deployment

TL;DR

This work investigates autonomous forest inventory using small legged robots to balance mobility and soil impact for under-canopy data collection. It introduces an end-to-end system combining LiDAR-Inertial Odometry, LiDAR-SLAM, dense local mapping, terrain modeling, a multi-level autonomy stack, and an online forest analysis pipeline that segments trees and estimates traits in real time. Seven field missions across Evo (Finland) and UK forests demonstrate autonomous operation and integration with real-time inventory outputs, including a 0.96 ha survey of the Forest of Dean with ~100 trees and 2 cm DBH accuracy. The study highlights both the feasibility of legged platforms for autonomous forest inventory and practical challenges (robustness, re-planning, safe maneuvering in clutter, and communication) to address for scalable deployment in real-world forestry contexts.

Abstract

We present a solution for autonomous forest inventory with a legged robotic platform. Compared to their wheeled and aerial counterparts, legged platforms offer an attractive balance of endurance and low soil impact for forest applications. In this paper, we present the complete system architecture of our forest inventory solution which includes state estimation, navigation, mission planning, and real-time tree segmentation and trait estimation. We present preliminary results for three campaigns in forests in Finland and the UK and summarize the main outcomes, lessons, and challenges. Our UK experiment at the Forest of Dean with the ANYmal D legged platform, achieved an autonomous survey of a 0.96 hectare plot in 20 min, identifying over 100 trees with typical DBH accuracy of 2 cm.

Figures (10)

• Figure 1: We present an autonomous forest inventory solution with legged platforms. Our system aims to autonomously drive a legged platform in an unknown forest plot, while recording environmental data used for tree segmentation and trait estimation.
• Figure 2: System overview. The Autonomy system executes the mission from a survey reference given by a human operator. State estimation provides a consistent scene representation, as well as dense clouds which are used for tree traits estimation. Forest analysis segments and estimates tree traits from point cloud data. The main output of the system is a forest inventory database with the main attributes of the surveyed forest.
• Figure 3: Survey Interface: GUI for the operator, implemented on RViz using interactive markers. The operator defines the area to be surveyed, while the survey plan is automatically proposed by the Mission Planner.
• Figure 4: Illustrative examples of the robot deployments during the different campaigns. (a) Coniferous forest in Evo, Finland. (b) Mixed forest in Wytham Woods, UK. (c) Oak forest in the Forest of Dean, UK.
• Figure 5: Summary of the seven survey missions executed across the three campaigns. We illustrate the periods where the robots operated fully autonomously, and where they required manual interventions from the safety operator.