Objectives
The CircuBot project aims to develop several enabling technologies (targeted in specific objectives – SO) that could significantly improve the waste sorting process modularly or combined within the finally targeted solution:
Specific objective 1 (SO1) – To enable versatile visual waste detection and sorting. CircuBot will develop an artificial intelligence-based system for waste sorting. In collaboration with industry partners, we will enlarge publicly available datasets and develop novel algorithms for the recognition and sorting of waste objects with respect to their visual appearance, dimensions, color, mass, damage levels, and material characteristics (to name a few waste sorting criteria).
Key performance indicators (KPIs): To enable reliable waste sorting in real-time, algorithms will be developed with respect to a certain speed (KPI1) – assessed by measuring the number of frames processed per–second and consequently the volume of objects that could be simultaneously processed on the moving conveyor for specified accuracy; object detection and classification accuracy (KPI2) – assessed by measuring Average Precision, Average Recall, and F1 score; and 3D object pose estimation accuracy (KPI3) using metrics that express deviation of referent geometry from the reconstructed ones (Dice coefficient, Chamfer distance, average 3D intersection over union).
Endpoint: Public data sets with waste images (PET/cans/electronic). Fully automatic recognition and classification system for waste tested on PET, cans, and electronic waste for robot-based waste handling and waste management monitoring and data analytics. By varying the volume of objects (or area of the conveyor), we will provide a report with tabular recommendations that will help companies to estimate their needs and select optimal workstation design.
Means of verification: The KPIs will be verified by comparison with related studies in scientific literature and human operators in the laboratory conditions that will be designed to reproduce conditions in the real waste sorting industry.
Specific objective 2 (SO2) – To enable cooperative human/robot waste picking: CircuBot will improve waste picking by optimizing: 1) robot grasping strategy for a typical class of objects, and 2) task sharing between humans and robots as cooperative agents in picking tasks in a safe and efficient manner. The SO2 will also include the evaluation of the success of catching with individual grippers for characteristic types of objects, as well as the analysis of the efficiency and safety of the work of the human operator, who collects individual types of waste on the conveyor belt.
Key performance indicators: Success of picking (KPI4) assessed by grasping/catching rate in % for three targeted types of grippers (customised gripper with vacuum suction cup, versatile soft gripper for handling various shapes and fragile objects, and qb soft hand human-like gripper for handling delicate objects) with regard to targeted waste materials (PET, cans, electronic waste); Efficiency of task scheduling in cooperative human-robot picking applications (KPI5) assessed by a number of picked items with reference to the same application with human operators only; Decrease in the cognitive workload of the human operator by at least 30% (KPI6) in human-robot cooperation mode that could be assessed by mBrainTrain smart headphones for non-invasive EEG recording or by participant survey.
Endpoint: Task-scheduling decision-making algorithm for human-robot cooperation for improved recycling performance, and guidelines for effective grasping/catching strategies for picking with regard to the waste type.
Means of verification: The KPIs will be verified in laboratory conditions (emulating picking the waste from the conveyor at a recycling site) for three different grippers mounted on a robot and different types of waste (KPI4), for robots as agents and for cooperative picking performed by both human and robot (KPI5, KPI6).
Specific objective 3 (SO3) – To improve the efficiency of the waste handling process on the robotics side: We will develop a novel strategy for dynamic pick-and-toss manipulation of objects to make compliant robot’s sorting more time and energy efficient – while ensuring assisting a human operator in a safe manner. We will exploit the principal advantages of cutting-edge compliant robot technology: 1) soft end-effectors will be used for precise catching of various types of waste, 2) the capability of storing/releasing energy will be used for achieving efficient throwing strategies, 3) the inherent compliance within robot joints will be used for enabling safe physical human-robot collaboration.
Key performance indicators: The proposed manipulation strategy will be evaluated through the following indicators: 1) time-efficiency (KPI7), i.e. time that passes between picking up the object and releasing the object from the robot’s end-effector will be reduced by at least 15%; 2) energy-efficiency (KPI8) will be increased by 10% and assessed by measuring the electrical current in robot joints and 3) accuracy of throwing will be higher than 95% (KPI9)– assessed by computing the percentage of thrown objects that reach the desired container.
Endpoint/Measurable outcomes: Pick-and-toss robotic strategy for increased workspace handling performance.
Means of verification: The proposed strategy will be verified in laboratory settings that emulate industrial conditions. The KPIs will be validated through the comparison between pick-and-toss and conventional pick-and-place strategies.
Specific objective 4 (SO4) – To digitalize the waste sorting management: We will significantly reduce the load to top-management and waste experts (whose physical presence is currently necessary to monitor sorting processes) by developing a cloud management platform that will be integrated with the CircuBot workstation. By using IoT, Big Data and Cloud computing we will enable more objective and transparent decision-making through the use of Data mining and Machine learning for pattern analysis, forecasting, fault detection and trend estimation.
Key performance indicators: The proposed tools will be developed to: 1) enable real-time collection of data from remote workstations (KPI10); 2) be platform-independent (KPI11) to be able to run on both mobile devices and PCs; 3) enable instant generation of reports (KPI12) from the collected data.
Endpoint: A cloud-based management platform will enable centralized monitoring, maintenance, and control of workstations. Smart data analytics will provide tools for waste counting, per-class analysis, reporting, and forecasting that could be intuitively accessed through the use of standard web browsers or mobile devices.
Means of verification: The platform will be released as an open-source project on the public GitHub repository, so that a wider audience could benefit from the project innovation outcome and adapt it for their use-cases or participate in further development.
Correspondence of the CircuBot project with objectives of the Green program
The CircuBot project fully corresponds with the defined general and specific objectives of the Green program of cooperation between science and industry by mobilizing the latest partner technologies and scientific excellence in applied AI (FINK) and collaborative robotics (SEE) for providing a modular and versatile intelligent system for enhancement of circular economy and waste management (FMGUB).
General objectives of the program: The CircuBot will develop a modular technology for versatile waste sorting in collaboration with three early adoptor companies: eReciklaza-Nis (electronics), Kappa Star Recycling (PET, cans, general communal waste), Unitrade Export (general communal waste). Therefore, the CircuBot proposes applied research aligned with the needs of Serbian companies that targets societal challenges (ecology and circular economy); The early adopters provided signed Letters of Support as commitment to participate in the evaluation of the CircuBot system in an industrial environment in the prospective second phase of the project to ensure the immediate industrial impact of the scientific outputs. The CircuBot also aims to create a team and establish scientific collaboration between the two research groups in the complementary areas of expertise AI and robotics, led by two promising young researchers with experience in managing large EU and national projects. Collaboration in robotics and AI/computer vision as complementary areas of expertise of the two partner institutions will create a skillful team and empower young researchers that can strengthen the scientific and innovation potentials of the institutions in vertical domains that fully meet priorities in national policies for research and innovations: the Strategy of Scientific and Technological Development of Serbia 2021-2025 “Power of Knowledge”, the Smart Specialization Strategy of Serbia (2020-2027), and Serbian Strategy for the Development of AI (2020–2025).
Specific objectives of the program: The project unequivocally contributes to achieving strategic national goals of zero pollution and a cleaner circular economy by improving the reusability of plastic (PET), metal (cans) and electronic waste as the biggest polluters of the environment. Reaching specific objectives of the project will be a building block towards a more efficient, intelligent, autonomous, and human-friendly waste sorting industry as a solid base for reaching the goals of “National strategy for waste management with the national waste management plan for the period 2020-2025”[1].
[1] Национална стратегија управљања отпадом са националним планом управљања отпадом за период 2020-2025. године.
