This project aims to address the challenges by developing: Deep learning-based solutions for versatile waste detection and sorting, a Robotic-based modular system for safe and efficient planning and cooperative handling of waste materials, and a Data-analytics tool to support and digitalize the waste management process planning. Within the scope of the project (Figure 1), CircuBot focuses on several specific objectives (SO) for advancing the recycling industry, with major implications to better reuse of waste materials, environmental protection, shortage of workers and enhancing human labor aspect in waste sorting and handling, and consequently creates added value in ecology, economy and healthcare.

Compared to the state-of-the-art, the CircuBot project has the following ambitions that we report as possibilities for significant advances in the related fields: 

1. Reduced workers’ overload through optimized scheduling of human-robot tasks. Contrary to existing solutions that aimed to completely replace human operators and resulted in limited success in industry practice, we propose the solution upon the compliant robots that operate side-by-side with humans. The solution will assist a human worker in the recycling process while reducing cognitive overload and enhancing the safety, speed and quality of sorting.
2. Increased reliability through the versatile picking of an object. Compared to the previous work^[1] which only uses vacuum suction cups for waste picking, this proposal aims to utilize the advantages of more sophisticated grippers, including a compliant robotic hand and three-fingered gripper, with the aim to achieve successful picking of various waste objects. We will perform comprehensive studies on which kind of grippers (customized 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) are the most suitable for different trash materials (plastic bottles, glass, metal cans, paper/cardboard). These novel results shall be presented at the international conference.
3. Increased robot efficiency with a pick-and-toss strategy. The previous research²⁰ explored the tossing of objects by focusing only on the robot’s position and velocity. In this proposal, we consider compliant robots whose impedance can vary in time. Therefore, our approach will be more comprehensive as we will be first to approach the tossing challenge from both position and impedance control viewpoints – with the aim to leverage the benefits of next-generation compliant robots in order to increase the efficiency, velocity and number of picks per hour. In a state-of-the-art, waste management strategy, a trivial pick-and-toss strategy has been employed to toss the waste, which is based on switching off the vacuum at the moment of throwing the object. Our novel approach for the waste pick-and-toss will provide tossing strategies for all three kinds of considered grippers – and it will eventually lead to the patent application.
4. Novel waste sorting dataset. The prerequisite for any application and/or advancement of the use of AI in some field is the availability of the relevant and sufficient amount of labeled data. One of the major reasons why there are no solutions to the considered problems is the lack of appropriate data sets (currently available data set contains only up-to 200 images per class). In collaboration with companies that signed letters of commitment to this project, we will yield the most comprehensive public data set that contains both images/videos of waste objects along with the accompanying sensors data measured with robotic arms (mass, inertia, stiffness) – and it will be available at the public GitHub repository.
5. Advanced waste sorting using Deep learning techniques. Previous studies on the topic were focused on applying state-of-the-art algorithms for 2D object detection and instance segmentation, which in industrial conditions suffers from high variability of waste types/classes, visual appearance, mass, and material characteristics (due to their overlapping and damage during the process of waste collection, transportation, and sorting). The proposed procedures will use an integral approach that fuses the assessment of the object’s 3D pose and shape (to estimate its damage and optimize robotic arm grasping) and the object’s mass and inertia computed from the robot arm. This will further enable us to perform the waste sorting concerning both visual and physical criteria – while increasing the reliability and safety of the grasping during the human-robot collaboration. This result will lead to the publication in a high-ranking international journal in AI. 
6. Digitalization of waste sorting management. Currently, waste sorting is among the least digitalized industries, which makes a series of challenges in both productivity and management. Through the utilization of IoT, Cloud, and Data analytics we will improve the monitoring and management of the waste sorting process. In this way, the dependence on human factors will be significantly reduced while the sorting process will be more transparent and easy to be managed. The platform will be released as an open-source project on the public GitHub repository, so that a wider audience could benefit from it. Also, the developed technology is foreseen for publishing in an international journal in the field of waste management.

Significance and potential for future extensions:

The proposed research is highly significant and applicable to the Program topic. Precisely, the proposed task-scheduling method will optimize the distribution of tasks between human operators and robots while maximizing the overall number of sorted items. A study on grippers will lead to an understanding of how to achieve a higher success rate in grasping. Pick-and-toss strategy has a high significance being a key enabler for the fast and safe sorting of waste.

A future extension envisions including EEG feedback, adding a robotic arm to make a dual-arm system capable of handling heavier objects (application in the construction waste management) and mounting a robot arm on a mobile platform to achieve larger flexibility of the overall system (application extends beyond the Program and contributes to the manufacturing and logistics).

[1] M. Koskinopoulou, F. Raptopoulos, G. Papadopoulos, N. Mavrakis and M. Maniadakis (2021) “Robotic Waste Sorting Technology: Toward a Vision-Based Categorization System for the Industrial Robotic Separation of Recyclable Waste,” in IEEE RAM, vol. 28, no. 2, pp. 50-60.