New Year, new beginning, good news comes in.

Congratulations to our colleagues Ji Shenghua and Tang Jinxun, whose co-supervised papers have been accepted by IEEE ICRA and IEEE ICIT respectively. The research questions in the papers all come from our team's thoughts and explorations on the application of low-speed unmanned driving scenarios, and propose our innovative solutions. Next, we will add luster and color to everyone's work with our new achievements.

①

A Point-to-distribution Degeneracy Detection Factor for LiDAR SLAM using Local Geometric Models

The IEEE International Conference on Robotics and Automation (IEEE ICRA, abbreviated as IEEE ICRA) is the top international academic conference in the field of robotics, ranking first in terms of scale and influence. It is also the flagship conference of the IEEE Robotics and Automation Society, aiming to provide an international platform every year for robot experts and scholars to present and exchange their research findings. The papers of IEEE ICRA are highly recognized by the industry, with strict reviews and acceptance criteria, so it is a very happy thing for robot scholars and experts to have their papers published in ICRA. IEEE ICRA 2024 (https://2024.ieee-icra.org/) will be held from May 13 to May 17 in Yokohama, Japan, and will gather researchers, students, and industrial partners from all over the world to discuss new innovations and breakthroughs, emphasizing the role of robotics and automation in addressing global challenges.

·Background

Lidar has good long-range data measurement capabilities and stability, and is one of the commonly used main sensors for robots to achieve simultaneous localization and mapping (SLAM). To enable robots to obtain precise position information, minimizing the geometric features of matching consecutive frames is the mainstream method at present. However, the relevant algorithms mainly rely on the geometric information obtained from lidar detection, so high-precision positioning can only be achieved in scenes with obvious geometric features.

Mapping effect under normal scenarios

However, due to the limitations of the working principle and physical characteristics of LiDAR, certain special scenes (such as long corridors, tunnels, and open areas) lack constraints on geometric information in some directions. As a result, the position estimation accuracy of the LiDAR SLAM system in the corresponding direction will be significantly reduced, causing the LiDAR SLAM system to be unable to accurately obtain the robot's position information. This phenomenon is called LiDAR SLAM degradation. If the robot continues to move without accurate position information, there will be significant safety hazards for both the robot and the environment.

Long corridor degradation scene

Empty degenerated scene

·Contribution

To address the issue that the current laser SLAM degradation detection methods still have significant errors, the paper proposes a point-to-distribution degradation detection method based on a local geometric model, which can judge the positioning reliability of the robot, thereby improving the robustness of the robot's operation.

·Principle

This method first divides the laser point cloud into multiple local point clouds based on the point cloud distribution density, then calculates the distribution probability of the current frame point cloud in the corresponding local point cloud space of the previous frame, and finally judges the degradation state of the robot by comparing the changes in the distribution probability between frames. The basic principle is shown in the figure below.

Degeneration factor

·Effectiveness

We use standard deviation to evaluate the robustness of the degradation detection (the lower the value, the lower the false detection rate). As can be seen from the following figure, the method we propose has better robustness in all the test datasets!

Robustness test

In addition, in the comparison experiment of false detection times, our method also has fewer false detection times.。

False detection test

②

Collision-free Edge-following Path Planner based on Adaptive Sampling Continuous Collision Detection

The IEEE International Conference on Industrial Technology (IEEE Industrial Technology International Academic Conference, abbreviated as IEEE ICIT) is one of the flagship conferences of the IEEE Industrial Electronics Society, dedicated to disseminating new ideas, research, and advancements in the fields of intelligent and computer control systems, robotics, factory communication and automation, flexible manufacturing, data acquisition and signal processing, visual systems, and power electronics. The 25th IEEE ICIT (https://icit2024.ieee-ies.org/) will be held from March 25 to 27, 2024, in Bristol, UK.

·Background

Outdoor tasks such as sweeping, logistics distribution, and watering and spraying plants require robots to have excellent edge-following/edge-approaching movement capabilities. However, when robots encounter obstacles during edge-following/edge-approaching movement, they face the contradiction of needing to be as close to the edge/obstacle as possible while also ensuring that they do not collide, thus posing challenges to rapid and accurate collision detection and path planning.

Real-time performance and detection accuracy are two core indicators of collision detection algorithms. Current methods find it difficult to balance both: either the collision detection is fast but with low accuracy, posing safety risks; or the accuracy is high but the timeliness is low, failing to meet industrial application requirements. In addition, conventional collision-free path planning based on B-spline curves usually does not consider the "cross-boundary" issue at the "entry point" and "exit point," while edge collision-free path planning must take this into account.

The lawn scene goes around the obstacle along the edge.

·Contribution 1

To resolve the contradiction between the real-time performance and detection accuracy of collision detection, the paper proposes Error-Limited Adaptive Sampling Continuous Collision Detection (ELAS-CCD), establishes a relationship between permissible error and sampling frequency, and achieves efficient collision detection with specified accuracy by inserting discrete footprints.

This method first discretizes the path, constructs a model between the allowable error between two frames and the insertion footprint frequency, then specifies the allowable error, calculates the insertion footprint frequency through the Newton iteration method, and finally performs collision detection.。

ELAS-CCD Schematic diagram

Track preprocessing

·Effectiveness 1

A visual comparison of various collision detection methods

Statistics of time consumption for various collision detection methods

Compared with the other three methods, the ELAS-CCD method proposed by us has improved in terms of detection accuracy and comprehensive time-consuming indicators, and the detection accuracy can be set, making it more suitable for practical applications.。

·Contribution 2

To improve the edge-encircling obstacle avoidance effect, the paper proposes an edge-encircling non-collision path planner EF-Planner based on B-spline curves, which integrates ELAS-CCD and constructs an edge-penalty term to ensure edge-encircling obstacle avoidance.

·Main principle

（a）Divide the collision segments and non-collision segments on the tracking path; (b) extract the gradient information of the road sections; (c) construct a multi-objective function for iterative optimization and solution.

Edge-Planner Schematic diagram

·Effectiveness 2

 

Iron railing scene along the edge around the obstacle

Compared to the baseline method Ego-Planner, the Edge-Planner we propose is closer to obstacles and edge objects when sweeping around obstacles, demonstrating superior sweeping efficiency and obstacle avoidance effects.

Compared to the baseline method Ego-Planner, the Edge-Planner we propose is closer to obstacles and edge objects when navigating around them, demonstrating higher cleaning efficiency and obstacle avoidance effects.

Thanks to the continuous deep work of countless scientific researchers, the industry has been able to thrive with numerous ships racing and thousands of sails competing. JT-Innovation values technological innovation, encouraging employees to dare to tackle tough challenges. We will persistently accumulate technical confidence day by day, providing and creating reliable solutions for the industry.