The RoboCup@Home league aims to develop service and assistive robot technology with high relevance for future personal domestic applications. It is the largest international annual competition for autonomous service robots and is part of the RoboCup initiative. A set of benchmark tests is used to evaluate the robots' abilities and performance in a realistic non-standardized home environment setting. Focus lies on the following domains but is not limited to: Human-Robot-Interaction and Cooperation, Navigation and Mapping in dynamic environments, Computer Vision and Object Recognition under natural light conditions, Object Manipulation, Adaptive Behaviors, Behavior Integration, Ambient Intelligence, Standardization and System Integration.
Fitting into robotic object manipulation, the project aims at evaluating the feasibility of using unconventional prehensility appendage as end effector. Nowadays, universal grippers seem to offer great alternatives for enabling robotic arm to grasp and hold objects. In order to refine this technology to our objectives experiments were conducted . The resulting device was implemented on an Universal Robot 10.
Prototyping and developing our version of an universal gripper was made possible with Eirlab resources. These resources mainly consisted of rapid prototyping equipment.
- 3d printer:
The support cup part of the universal gripper was produced using 3d printers. 3d models were done in Blender then sliced with Cura. The gcode was then printed on Ultimaker printer following Eirlab recommandations.
- Laser cutting machine:
A laser cutting machine was used to produce a linking plate between the robot arm and the gripper. The cutting pattern was defined with Inkscape and cut following the machine instructions.
In order to achieve robotic object manipulation both a robot arm and a end effector are needed. An UR10 was available at ENSEIRB-MATMECA and saved us time needed to work on the universal gripper.
The UR10 is manufactured by the dannish company Universal Robots . This robot arm has a polyscope graphical user interface on 12 inch touchscreen with mounting and all the movements can be made on a 360° angle (6 degrees of freedom).
At first we tried using ROS to control the arm but due to its complexity and not needed functionalities we resigned to using the built-in touch screen controller. The project being Indeed mainly a proof of concept for the gripper, not controlling extensively the arm with ROS was a time sparing choice.
Tasks that appear simple to humans, such as picking up objects of varying shapes, can be extremely complicated for robots. The standard approach is based on a hand design with two or more fingers. It involves to be careful about the fragility of the object and so force sensors at the fingertips are often used. This approach has many advantages but requires a complex brain which can make a multitude of decisions.
The universal gripper approach replaces individual fingers by a material which is able to mold itself around the object. It is called universal because it is adapted to many shapes. This process reduces the number of elements to be controlled and therefore can have advantages in terms of cost and gripping speed.
We have chosen to focus on a universal gripper in its simplest form. Which means a simple latex balloon filled with a granular matter (such as coffee, silica beads or polystyrene beads). When the gripper first touches an object, the beads are free to flow inside the bag. Pushed onto the object, the gripper partially envelops it. The grip is secured when air is sucked out of the bag, compressing the beads so close to each other that they no longer can move.
The development of the cup went through different phases. At first we experimented on a smaller scale. Making smaller cup and smaller gripping part enabled us to refine our prototypes.
Our first prototype was of cone shape. This shape seemed not appropriate for the balloon. Moreover its top opening was too narrow to fit the the rubber balloon tail and had sharp edges that could have popped the balloon. A more rounded support was then made to better adapt to the roundness of the gripping part.
For our final prototype we have flatten the cup to again better fit the balloon. We added some ridges inside the cup to enable the gripping part to grip on the support while grabbing on object. Before all the weight of the lifted object was supported by the balloon’s tail and with heavier objects it could have tore. The cup was also larger to accommodate for larger balloons. However the ridges were not steep and large enough to lift heavy objects, the ballon was not gripping to those protuberances.
Better result in heavy lifting could be achieved by enlarging those ridges or by reducing the circumference of the cup. Indeed a smaller cup could be pushed in the gripping part and the balloon could solidify around it.
To reach vacuum in the balloon we were using a vacuum cleaner. It was then necessary to make a junction piece serving as a link between the support cup tube and the cleaner's one.
Our solution done it was then needed to mount it on the UR10. A mounting plate was made to deliver this role.
Vacuum cleaner specs
|Air flow||59 l/s|
|Depression||227 / 22.7 mbar|
|Max absorbed power||1500 W|
We tested filling the universal gripper with polystyrene beads then with coffee. We then lifted various objects differing in sizes, weights and matter compositions. We used a minimalist approach to run the tests, leaving aside all functionalities that were not needed for their courses. The gripper was positioned on an object then lifted by human handling. The vacuum cleaner and the support cup were also held together by hand. Before each lift the target object was weighted. Such experiments were conducted to determine the weight and shape gripping limits.
The first test was on a balloon filled with polystyrene beads. The system was able to lift the following objects:
|cissors||metal and plastic||25|
However the gripper started to face problems with intricate shapes, shapes that did not offer enough contact surface (metal bar, cylinder of 145g, 1324g), and with weights of around 2kg and more (lifting weigth, grinder of 2kg, 2.239kg). Lifting was not consistent and often required presenting the objects in a particular manner to the gripper. With really heavy and flat objects the lifting could not be achieved and they slipped out of grasp.
For the next experiment we filled the gripper with ground coffee. But we quickly figured out that was not working as well as with polystyrene. Coffee seemed to be too heavy and too dense imposing its own constraints on the gripper. There was not as much free air in the ballon due to finer grain. It was much more challenging for the gripper to mold itself and relieve objects. When relieving an object the air had to find its way around a much more compact and dense matter. However we still managed to observe the same kind of results only with harder time to obtain.
The following table summarize the difference between the two solutions:
|Filled with polystyrene||filled with coffe|
|Cost per gripper (€)||2 (5L 6€)||10 (3 coffee packets)|
|time between two usage (s)||2-3||10|
As a result of our experimentations polystyrene beads seems to be better suited for the task. However the gripper appeared to be more effective while not filled completely. The beads should be able to move freely in the gripper before reaching vacuum. When upscaling the solution we found out that the size does not make a difference in the power needed to suck the air out and the gripping ability. But larger balloons enabled larger object lifting with much more safety mainly due to a larger area of contact.
As a full scale proof of concept we mounted our final solution on the UR10. A pump not being available we had no other choice but to stick with the vacuum cleaner used during testing. If we had accessed a pump we could have used a microcontroller or an Arduino for automation. It could then have been paired with the robotic arm into a program using ROS.
For the demo we used the tablet to create waypoints to reproduce the kind of movement that a homebrew software could have enabled. During a recorded movement the vacuum cleaner is activated and shut manually. 4 waypoints were enough to grab an object displace it then relieve it.