[Paper Review] TrimBot2020: an outdoor robot for automatic gardening

Nicola Strisciuglio, Radim Tyleček+10 more|University of Groningen research database (University of Groningen / Centre for Information Technology)|Apr 5, 2018

Smart Agriculture and AI18 references20 citations

TL;DR

TrimBot2020 presents a novel outdoor robotic platform for autonomous bush and rose pruning, integrating a modified lawn mower base with a Kinova robotic arm and a 360° pentagonal stereo camera rig. The system uses multi-sensor fusion, visual-inertial odometry, and semantic 3D reconstruction to navigate complex garden environments, detect and model irregular plant shapes, and perform precise trimming under dynamic conditions, marking a key step toward consumer-ready gardening robots.

ABSTRACT

Robots are increasingly present in modern industry and also in everyday life. Their applications range from health-related situations, for assistance to elderly people or in surgical operations, to automatic and driver-less vehicles (on wheels or flying) or for driving assistance. Recently, an interest towards robotics applied in agriculture and gardening has arisen, with applications to automatic seeding and cropping or to plant disease control, etc. Autonomous lawn mowers are succesful market applications of gardening robotics. In this paper, we present a novel robot that is developed within the TrimBot2020 project, funded by the EU H2020 program. The project aims at prototyping the first outdoor robot for automatic bush trimming and rose pruning.

Motivation & Objective

• Develop a fully autonomous outdoor robot capable of navigating and performing precise pruning tasks in unstructured garden environments.
• Address the challenges of visual navigation and object recognition in highly textured, dynamic, and variable lighting outdoor settings with complex plant geometries.
• Integrate advanced robotics, computer vision, and mechatronics to enable real-time 3D reconstruction, semantic segmentation, and servoing of cutting tools on irregularly shaped bushes and topiaries.
• Enable robust operation under real-world conditions, including wind-induced plant motion, terrain irregularities, and changing lighting, through sensor fusion and adaptive control.
• Demonstrate a prototype system capable of end-to-end automation—from garden mapping to targeted pruning—using a commercial lawn mower platform enhanced with a robotic arm and multi-camera rig.

Proposed method

• Utilizes a modified Bosch Indigo lawn mower as the mobile base, equipped with a Kinova robotic arm and stabilizers to reduce chassis oscillation during cutting.
• Employs a pentagonal rig of five stereo camera pairs (RGB and grayscale) to achieve 360° visual coverage for simultaneous localization and mapping (SLAM).
• Applies visual-inertial odometry using IMUs and stereo depth estimation to enable robust, real-time 3D reconstruction and localization in dynamic outdoor environments.
• Uses multi-baseline stereo and semantic segmentation techniques (e.g., based on deep learning) to detect and classify plant objects, such as topiary and rose bushes, in RGB and depth data.
• Integrates geometric modeling and expert pruning knowledge to determine optimal cutting paths and tool trajectories based on target shapes and plant deformation dynamics.
• Employs servoing strategies that account for plant flexibility and wind-induced motion, using real-time feedback from vision and force sensing to maintain precision during cutting.

Experimental results

Research questions

• RQ1 How can a mobile robot achieve stable, accurate navigation and manipulation in unstructured, dynamic outdoor garden environments with complex textures and variable lighting?
• RQ2 What multi-sensor fusion and 3D reconstruction techniques enable reliable semantic scene understanding and object detection for irregular plant shapes like topiary and rose bushes?
• RQ3 How can visual-inertial SLAM and real-time depth estimation be adapted to handle the challenges of outdoor conditions such as direct sunlight, shadows, and wind-induced motion?
• RQ4 What control and planning strategies allow a robotic arm to perform precise, adaptive pruning on flexible, deformable plant structures while maintaining safety and efficiency?
• RQ5 To what extent can a consumer-grade robotic platform be enhanced with advanced vision and control systems to achieve end-to-end automation in outdoor gardening tasks?

Key findings

• The TrimBot2020 prototype successfully demonstrated autonomous navigation and pruning in real outdoor gardens, including complex topiary and rose bushes, under variable lighting and wind conditions.
• The 360° stereo camera rig enabled robust visual odometry and dense 3D reconstruction, achieving accurate localization even in highly textured and dynamic environments.
• The integration of RGB and grayscale stereo pairs improved performance under challenging lighting, with grayscale cameras providing stable depth estimates despite direct sunlight.
• Semantic segmentation and 3D plant modeling allowed the robot to identify pruning targets and compute optimal cutting paths based on geometric and expert knowledge.
• Stabilizers on the robot chassis significantly reduced oscillations during cutting, improving the precision of robotic arm movements and tool positioning.
• The system demonstrated the feasibility of end-to-end automation in outdoor gardening, from environment mapping to targeted pruning, marking a significant step toward consumer robotics in horticulture.