Periodic Reporting for period 2 - ReconCycle (Self-reconfiguration of a robotic workcell for the recycling of electronic waste)
Reporting period: 2021-04-01 to 2022-09-30
Currently electro-recycling is still mostly performed following the crude “crush-and-separate” method, where devices are crushed and split into tiny particles, which are then physio-chemically separated into reusable components. Many times, this requires manual work to remove dangerous components or to disassemble sub-parts that cannot go into the crusher. Recycling companies are here faced with tons of material that arrive in different states of damage. The manual pre-processing of this kind constitutes a major (cost-)effort. An inappropriate “alternative” for this is currently still to ship waste to developing countries and abuse their cheap labour to address these problems (or let the waste just be dumped there). Clearly, there is a very strong incentive to improve on this and this way increase profitability, efficiency, and positive environmental impact of recycling and, thus, even partial automation of some of these steps can make a big difference.
One specificity of electronic recycling is that usually many different models of the same type but in vastly different conditions need to be disassembled. Thus, in this domain we have to deal with large batches of the same type of device, which limits variability at least to some degree, but we are at the same time confronted with many different models within any given device-type and the fact that they are in vastly different conditions, which requires a flexible robotic system that adapts to these circumstances. As a consequence, electronic waste recycling offers an ideal test-bed for self-reconfiguration of robotic cells as the complexity of the required changes within a given device-type is not too high.
The aim of the project is to introduce self-reconfigurable hardware and software for disassembly of electronic devices. A two-step procedure is foreseen:
a) When changing from one device-type to another, the reconfiguration shall be performed in an interactive mode, where the application-engineer will be able to provide his/her input.
b) When changing from one device-model to another within a given device-type, the cell shall perform re-configuration (predominantly) on its own through a combination of sensorimotor learning approaches and other AI techniques.
The challenge of this is to provide methodologies for re-design of the recycling cell and approaches for fast re-programming, re-use and adaptation of manipulation actions for disassembly with soft robots and grippers.
One specificity of electronic recycling is that usually many different models of the same type but in vastly different conditions need to be disassembled. Thus, in this domain we have to deal with large batches of the same type of device, which limits variability at least to some degree, but we are at the same time confronted with many different models within any given device-type and the fact that they are in vastly different conditions, which requires a flexible robotic system that adapts to these circumstances. As a consequence, electronic waste recycling offers an ideal test-bed for self-reconfiguration of robotic cells as the complexity of the required changes within a given device-type is not too high.
The aim of the project is to introduce self-reconfigurable hardware and software for disassembly of electronic devices. A two-step procedure is foreseen:
a) When changing from one device-type to another, the reconfiguration shall be performed in an interactive mode, where the application-engineer will be able to provide his/her input.
b) When changing from one device-model to another within a given device-type, the cell shall perform re-configuration (predominantly) on its own through a combination of sensorimotor learning approaches and other AI techniques.
The challenge of this is to provide methodologies for re-design of the recycling cell and approaches for fast re-programming, re-use and adaptation of manipulation actions for disassembly with soft robots and grippers.
The current focus of the project is to implement reconfiguration and adaptation procedures to move away from the initial archetypical solution suitable for dismantling of a specific device model to a more general solution that can deal with multiple models of the same type, e.g. different models of heat cost allocators (HCAs). For this purpose we developed procedures to generate more general disassembly protocols that can map the current situation in the disassembly process, where information is primarily extracted by vision sensing, to the appropriate actions of the cell. To enable the dismantling of different devices of the same type, we also investigated new adaptation and reconfiguration procedures to compute the appropriate parameters for the desired disassembly actions and to appropriately change the design of the workcell when dismantling different HCA models.
To create a flexible environment in which such processes can be implemented, we built upon the modular hardware and software hardware architecture developed in the first reporting period. New developments included the design of new adaptable vise and cutter modules that provide sufficient flexibility to process a variety of devices of the same type. On the software side of our works, we implemented additional functionalities in the ReconCycle ROS-based environment for programming & workcell control. To provide the cell with the flexibility needed for the effective integration of advanced procedures required by disassembly solutions with adaptation and self-reconfiguration capabilities, we developed several new FlexBE behaviors. Most importantly, we implemented a unified FlexBE protocol that can be used for automated disassembly of different heat cost allocator models.
The important current research focus is the development of an Action Predictor module, which purpose is to predict the next action to be taken based on the sequence of previously taken actions and the current state of the workcell as sensed by vision. The action predictor module is based on the parts relationship graph used to determine which actions should be taken next. The precise actions are defined within the Context Action Framework, which defines the required disassembly actions and the related processes (cutting, levering out, moving, pushing, turning over, etc.).
We continued working on unifying motion trajectories and force profiles in order to provide more adaptable and flexible tactile manipulation capabilities under such circumstances. Tactile skills are achieved by combining three basic motion primitives: contact initiation, manipulation, and contact termination. We also worked on the further developments of SoftHand 2 and Variable Stifness (VS) Gripper to increase their capabilities in terms of grasping in the recycling domain. New tool exchange systems for SoftHand 2 and VS Gripper were constructed and we started to work on the sensorization of SoftHand 2 to enable the reconstruction of its posture without relying on vision. Finally, key performance indicators – both for specific building blocks of the cell as well as for the ReconCycle use-cases – were identified.
Up to now 14 scientific papers have been published at scientific conferences and journals. The ReconCycle systems were demonstrated at industrial fairs such as EMO 2021 and BI-MU 2022.
To create a flexible environment in which such processes can be implemented, we built upon the modular hardware and software hardware architecture developed in the first reporting period. New developments included the design of new adaptable vise and cutter modules that provide sufficient flexibility to process a variety of devices of the same type. On the software side of our works, we implemented additional functionalities in the ReconCycle ROS-based environment for programming & workcell control. To provide the cell with the flexibility needed for the effective integration of advanced procedures required by disassembly solutions with adaptation and self-reconfiguration capabilities, we developed several new FlexBE behaviors. Most importantly, we implemented a unified FlexBE protocol that can be used for automated disassembly of different heat cost allocator models.
The important current research focus is the development of an Action Predictor module, which purpose is to predict the next action to be taken based on the sequence of previously taken actions and the current state of the workcell as sensed by vision. The action predictor module is based on the parts relationship graph used to determine which actions should be taken next. The precise actions are defined within the Context Action Framework, which defines the required disassembly actions and the related processes (cutting, levering out, moving, pushing, turning over, etc.).
We continued working on unifying motion trajectories and force profiles in order to provide more adaptable and flexible tactile manipulation capabilities under such circumstances. Tactile skills are achieved by combining three basic motion primitives: contact initiation, manipulation, and contact termination. We also worked on the further developments of SoftHand 2 and Variable Stifness (VS) Gripper to increase their capabilities in terms of grasping in the recycling domain. New tool exchange systems for SoftHand 2 and VS Gripper were constructed and we started to work on the sensorization of SoftHand 2 to enable the reconstruction of its posture without relying on vision. Finally, key performance indicators – both for specific building blocks of the cell as well as for the ReconCycle use-cases – were identified.
Up to now 14 scientific papers have been published at scientific conferences and journals. The ReconCycle systems were demonstrated at industrial fairs such as EMO 2021 and BI-MU 2022.
Progress beyond the state of the art can already been observed in several areas of ReconCycle developments. A new hardware and software architecture for recycling cells has been developed, with support for modularity and reconfigurability both in hardware and software. This should enable the disassembly of a variety of devices and models for recycling purposes for the first time. Progress beyond the state of the art has been shown also in the design of soft hands, tools, and grippers suitable for disassembly of electronic devices, which are important for the area of soft robotics in general.
To further the developments of recycling technologies in the robotics research community, ReconCycle supported the Robothon competition organized at Automatica 2021. The development of qb SoftHand 2 Research culminated in a coordinated technology transfer action among the partners and introduction of a new product to the market at ICRA 2022.
By the end of the project we expect to develop partially automated methodologies for re-design of the recycling cell to enable switching of the disassembly procedure from one electronic device to another.
From the application point of view, ReconCycle aims to substantially reduce human effort and increase accuracy and efficiency of recycling. As ReconCycle uses highly compliant robots, humans will be able to operate together with the machines to complete the missing steps. This reduces the automation complexity further and brings this project into a feasible regime.
To further the developments of recycling technologies in the robotics research community, ReconCycle supported the Robothon competition organized at Automatica 2021. The development of qb SoftHand 2 Research culminated in a coordinated technology transfer action among the partners and introduction of a new product to the market at ICRA 2022.
By the end of the project we expect to develop partially automated methodologies for re-design of the recycling cell to enable switching of the disassembly procedure from one electronic device to another.
From the application point of view, ReconCycle aims to substantially reduce human effort and increase accuracy and efficiency of recycling. As ReconCycle uses highly compliant robots, humans will be able to operate together with the machines to complete the missing steps. This reduces the automation complexity further and brings this project into a feasible regime.