Project Description
Manufacturing industries often deal with items of different shapes, sizes and weights, requiring pick-and-place activities to be carried out.
Usually, human operators carry out the picking activity, as model-based approaches often struggle when dealing the variety of items to be collected. If one were to automate the process, Artificial Intelligence (AI) can be used to address the issue.
Previous activities at IMA S.p.a. involved the design of AI-based vision algorithms to determine optimal suction points on a variety of items, showing remarkable results when dealing with irregular shapes. Nevertheless, only relying on AI still leads to suboptimal results in many common scenarios.
The project aims at improving the robustness of the current algorithm, comparing different training dataset for the AI-based solution and implementing additional heuristics.
Activities will include a first study of AI for computer vision algorithms, which will lead to the definition of evaluation criteria for the already available libraries. After the first evaluation step, possible ways to improve the system will be researched and implemented. Lastly, the algorithm will be tested on a real ROS-controlled platform to assess its performance.
Prerequisites
- Strong initiative and attitude to problem solving
- Basic knowledge and strong interest in software programming
- Basic knowledge of robot kinematics
Not required but good to have
- Sound knowledge of programming techniques
- Knowledge of python programming language
- Basic knowledge of computer vision
Target Objectives
- Acquire competence in AI for computer vision
- Strengthen the competence in software programming
- Acquire competence in ROS operating system
Expected Results
- Development of an evaluation benchmark for vision algorithms
- Design of an improved suction detection algorithm
- Comparison of the new algorithm with the current algorithm
Industrial Robotics, Mechatronics & Automation Laboratory @ Bologna 