Robot recycling detects metal, paper and plastic
A robotic system has proven to have a high level of accuracy at identifying different types of materials, to assist refuse sorting.
Called RoCycle, the bot proved to have 85% accuracy at identifying stationary objects and 63% for those on a moving belt.
Developed by a team from MIT, USA, and Yale University, the electric soft robotic system consists of a gripper made of two-fingered handed-shearing auxetic (HSA) cylinders with a pressure sensor on the inner and strain sensor on the outer sides, covered in a haptic sensing glove.
The system is motor-driven and the gripper actuated by a servo, which enables greater control and scalability than fluid-driven actuators. The HSA cylinders counterrotate against each other, and the level of deformation they undergo as well as the grip width change from open to holding the object, helps determine characteristics including size and softness or stiffness.
As the sensors are conductive, they can measure changes in electrical signals to identify if a material is metal.
To set a baseline, the team recorded values for open and closed gripper rates, and then compared these against data from when holding objects, including a magic 8 ball, a deflated football, a foam brick, and a whiteboard eraser. Combining data from the various points establishes how much force is required to hold the object, which helps identify the material.
‘After the gripper closed, the pressure and strain values are recorded. Each object is then picked up and shaken to determine robustness of grasp hold. We recorded data before shaking as the extra motion caused shifts in the grasping normal forces, changing the pressure sensor values dramatically,’ the paper reads.
RoCycle has proven effective thus far, but has many areas to improve upon. While the machine has a high rate of success with simple objects, such as recognising the difference between paper and plastic, layered or coated materials present a problem as it can only identify the outer material.
Now, the team plans to couple the machine to a camera-based system to improve accuracy and sorting speed.
The project was carried out under the MIT Computer Science and Artificial Intelligence Lab programme, with funding from Amazon, JD, the Toyota Research Institute and the US National Science Foundation.
You can read the full paper here: bit.ly/2DdL4ZJ