TL;DR
This paper tackles the challenge of precise navigation for pure electric driverless crawler construction machinery, especially in environments with weak GPS signals. They achieve this by fusing LiDAR-based SLAM with an improved pure pursuit algorithm, enabling robust path tracking and laying groundwork for autonomous construction.
What Problem Does This Paper Address?
Construction machinery operates in harsh, often GPS-denied environments (like tunnels or under structures), making traditional GPS-reliant autonomous navigation impossible. Furthermore, the unique kinematic characteristics of crawler-based machinery (unlike wheeled vehicles) necessitate specialized control algorithms. Existing driverless solutions for cars are not directly applicable, and research in this area for construction machinery is still nascent, often relying on remote control. This paper aims to bridge this gap by developing a robust, purely vision- and sensor-based navigation system for pure electric crawler construction machinery.
Methods
The core of the solution lies in two key components:
- Fusion SLAM based on Improved NDT: To overcome GPS limitations, the authors employ a Simultaneous Localization and Mapping (SLAM) system. They use a Normal Distributions Transform (NDT) algorithm, enhanced with a Levenberg-Marquardt iterative method for improved efficiency and robustness. This NDT approach uses LiDAR to build a point cloud map of the environment and then matches real-time scans to this map for precise localization. Crucially, it's fused with IMU and odometer data to provide initial pose estimates, accelerating the registration process and improving accuracy, especially in challenging conditions.
- Improved Pure Pursuit Algorithm: Recognizing that standard pure pursuit algorithms are designed for Ackerman steering (like cars), the authors adapt it for crawler construction machinery. This involves modifying the motion model to account for the differential drive system of crawlers, where steering is achieved by controlling the speed of individual tracks. This allows the machine to execute turns and follow paths by adjusting the speeds of its left and right tracks.
Datasets
Not discussed in the paper. The authors built their own simulation platform and conducted real-vehicle tests using their custom-built electric crawler excavator.
Key Insights and Ideas
- NDT for Robust Localization in GPS-Denied Areas: The NDT-based SLAM, augmented with IMU/odometer fusion, proves effective for accurate positioning even without GPS, a critical requirement for construction sites.
- Adapting Pure Pursuit for Crawlers: The modification of the pure pursuit algorithm to handle differential drive kinematics is a key innovation, enabling precise path following for this specific type of machinery.
- Integrated System for Practical Application: The paper demonstrates the successful integration of these two components into a functional system, validated through both simulation and real-world tests.
Key Results
- Mapping Accuracy: The fusion SLAM method achieved high mapping accuracy in small- and medium-sized scenes, with maximum errors in the forward and lateral directions of 0.05m and 0.06m respectively, and root mean squared errors of 0.029m and 0.037m.
- Positioning Accuracy: The fusion SLAM positioning track was found to be consistent with the reference track (RTK), with maximum forward and lateral errors of 0.065m and 0.06m, and root mean squared errors of 0.04m and 0.032m, respectively.
- Simulation Performance: In simulation, the driving track closely matched the target path, with maximum forward and lateral errors of 0.082m and 0.162m, and root mean squared errors of 0.025m and 0.055m.
- Real Vehicle Test Performance: The real vehicle test showed similar promising results, with maximum forward and lateral errors of 0.083m and 0.152m, and root mean squared errors of 0.031m and 0.052m, demonstrating the feasibility of the proposed method.
Limitations and Future Work
- Scale of Scenes: The current fusion SLAM method, using basic NDT, is primarily effective for small- to medium-sized scenes. Future work will focus on improving the NDT registration for large-scale environments.
- Speed Limitations: The improved pure pursuit algorithm is currently suited for low-speed mobile construction machinery. Future research will explore developing algorithms for high-speed machinery and incorporating vehicle dynamics.
- Path Planning: The current system focuses on trajectory tracking. Future work will integrate perception (using PointPillars and CenterPoint) and cost maps with algorithms like Hybrid A* for autonomous path planning and decision-making, tailored to construction machinery operating conditions.
Summary updated