Automatic Robot Path Planning for Visual Inspection from Object Shape (2312.02603v1)

Abstract: Visual inspection is a crucial yet time-consuming task across various industries. Numerous established methods employ machine learning in inspection tasks, necessitating specific training data that includes predefined inspection poses and training images essential for the training of models. The acquisition of such data and their integration into an inspection framework is challenging due to the variety in objects and scenes involved and due to additional bottlenecks caused by the manual collection of training data by humans, thereby hindering the automation of visual inspection across diverse domains. This work proposes a solution for automatic path planning using a single depth camera mounted on a robot manipulator. Point clouds obtained from the depth images are processed and filtered to extract object profiles and transformed to inspection target paths for the robot end-effector. The approach relies on the geometry of the object and generates an inspection path that follows the shape normal to the surface. Depending on the object size and shape, inspection paths can be defined as single or multi-path plans. Results are demonstrated in both simulated and real-world environments, yielding promising inspection paths for objects with varying sizes and shapes. Code and video are open-source available at: https://github.com/CuriousLad1000/Auto-Path-Planner

• The paper presents a novel methodology that automatically generates inspection paths by processing 3D point cloud data of object surfaces.
• The system integrates a depth camera with a robot manipulator to extract and filter geometric profiles from objects of varying shapes and sizes.
• Tests in both simulated and real-world environments demonstrate efficient path planning, reducing reliance on manual trajectory design.

Introduction to Robot Path Planning

Robot path planning plays an integral role in the field of automation and assists in enhancing the efficiency and safety of tasks such as visual inspection. Traditional methods of visual inspection, predominantly performed by humans, are often time-consuming and potentially hazardous, requiring the presence of a person in unsafe environments. Advances in AI and robotics have paved the way for more autonomous solutions in various industries, notably influencing the procedures associated with quality control and maintenance tasks.

Automating Visual Inspection

The paper presents a novel methodology for automating robot path planning for visual inspection purposes. The system integrates a single depth camera mounted onto a robot manipulator to capture 3D data in the form of point clouds, which represent the surface details of objects. These point clouds are then processed and filtered to extract the object's profile, regardless of its shape or size.

This innovative approach leverages the geometry of the objects to generate inspection paths that follow their shape. The methodology allows both single and multiple inspection path plans to adapt to objects with different dimensions and geometric complexities. The paper's approach takes steps to address a crucial issue in traditional visual inspection methods—dependence on training data, which often requires labor-intensive gathering and is hindered by variability in objects' sizes and shapes.

Methodology and Results

The proposed system operates by capturing multiple point clouds, refining the point cloud data through filtering and clustering, and deriving normals to determine the robot camera's optimal positioning. The collected point cloud data is processed to identify the cluster most representative of the object profile. From this profile, target poses are generated, guiding the robot manipulator along the calculated inspection path.

The methodology was tested with both simulated and real-world objects, demonstrating that it could successfully generate appropriate path plans for objects of various shapes and sizes. In experiments involving complex objects like car models and tunnels in virtual environments as well as physical items like benches and suction rolls, the system proved to be effective in path planning without the need for manual data collection or predefined trajectories.

Conclusion and Discussion

In drawing conclusions, the research indicates the system's potential to automate visual inspection and 3D modeling without depending on CAD models, existing data sets, or human-prescribed paths. Although subject to certain limitations, such as the occasional appearance of "ghost" targets and residual points in the filtered point cloud, the system provides a malleable solution adaptable to different inspection requirements.

Furthermore, the paper discusses the implications of the number of point cloud samples and filtering thresholds on the quality and density of the target paths, offering insight into optimizing the system for various operational conditions. The authors also highlight their confidence in the open-source availability of their code and invite further advancement by the research community.

In conclusion, this approach to robot path planning embraces the fusion of 3D point cloud data and robotics to streamline the visual inspection process, which aims to shift repetitive and dangerous tasks from humans to robots, leading to safer and more efficient industrial operations.