After recovering from a very successful World Cup last summer, we are again busy with the development and improvement of our robots. Read below what we are up to!
MSL workshop
Tech United is going to organize the 11th MSL workshop together with VDL Robot Sports. The workshop will be held on the 16th and 17th of December in Eindhoven.
The theme of the workshop is “MSL the next decade”. It will mainly consist topics on how the league will develop in the next 10 years and how we can ensure that in the future enough teams will be motivated to join the league. Also, the workshop facilitates sharing of knowledge between new and already existing teams.
Portugal visits Tech United
At the end of September we had three team members from the Minho team from Portugal visit us for a full week. Because of the European Championship earlier this year in their city, they got motivated to restart their robot soccer team. Tech United played against Minho regularly until the team stopped in 2016.
During the visit in Eindhoven, they had the chance to speak to many of our team members and to get advice on several topics, among which: the processing of camera images, programming of tactics and synchronization between the robots. Minho will continue to work with the robots from 2016. They got advice from Sanders Machinebouw and us to improve their ball handling and driving.
Grasp detection
Our robot HERO currently has no method of determining whether it's holding something. In some situations this might be an issue. For instance, if the robot tries to bring something to you and is unable to grasp it, it will think it successfully grabbed the object and return empty-handed.
Our team member Rodrigo is currently trying to solve these problems. Right now, to determine if it's holding something, HERO measures the distance between its two fingers, squeezes them lightly, and measures the position a second time to compare it to the starting position. We call this method "active grasp detection".
The technique works flawlessly for objects like a can of soda, or an apple, but it can’t be used in all situations since it fails for objects like a soft sponge.
HERO needs full knowledge of the object that needs to be grasped, that is what makes this problem so challenging. Hopefully in the future HERO can know if he has something in his hands in every situation.
YOLOv7
Currently we are looking into YOLOv7 to improve object detection. YOLOv7 is an object detection algorithm, which makes use of a neural network. This neural network is first trained where it learns features of the classes. For now, we have trained the network on a dataset with 4 classes: coke, sprite, pringles and orange. Which means that the robot can recognize these objects. The network predicts whether there are objects in the image by extracting features from the image. When objects are detected, a bounding box is placed around it.
This algorithm is used when the robot has to pick up an object, for example when it needs to store the groceries or serve a drink. An example can be seen in the image, where the robot is asked to serve a coke. The coke is detected and classified with 89% confidence.
Swerve Drive
Next to the well-known TURTLE robots, the team started earlier this year with the development of the Swerve Drive platform. This platform has three wheel sets, where each set consists of a steering and driving motor that can be controlled independently. The main benefit of this system compared to the old omni-wheel system is that we can use standard tires, which could give us the opportunity to reach a higher acceleration, without losing agility. The first tests look promising, but in the next tournaments the new platform will really have the chance to prove itself. The platform also gives us the possibility in the future to drive outside or on (artificial) grass.
Trick shots
As you might have seen on our social media channels, we are working om some nice trick shots. We now have shot an apple off one’s head and a pylon off a bar. You can follow us and see our latest tricks on our YouTube channel, Instagram, Facebook and Twitter (see links below). Do you have any suggestions on what trick shots to do next? Let us know!
'
