News from the week
We discuss the YOLOv11 model release, Tesla's "We, Robot" event, and Petoi's Nybble Q robot
YOLOv11
The YOLOv11 model got released on September 30. YOLO (You Only Look Once) is a class of Machine Learning models that facilitate object detection and image classification with great speed and accuracy. YOLO is very popular in applications that require real-time object detection such as autonomous driving and robotics navigation. Our posts describing how to let a Vector robot detect another Vector robot (with the Python SDK) using the YOLO models have been very popular… in the past we have written on YOLOv4 vs YOLOv5, YOLOv5 vs YOLOv6 vs YOLOv7, YOLOv8, YOLOv9, and most recently YOLOv10.
YOLOv11 is the third YOLO model released this year, which shows the pace at which research tries to improve on the model. The general theme has been to try to shrink the model while maintaining the same or slightly better accuracy. Shrinking the model helps to improve the speed at which the model runs during inference, thus making object detection faster and useful for more real-time use cases. Shrinking the model also saves memory which is useful for deploying the model in edge devices such as robots which are memory constrained. YOLOv11 is about 22% smaller than YOLOv8 (both models are built by the same team based on similar concepts)
Like our previous posts, we fine-tuned the YOLOv11 model to help the Vector robot detect another Vector robot. We have a notebook which you can use train the YOLOv11 model on a Google Colab instance with the Vector dataset (which we have made publicly available at Roboflow). Our results have not been good… at least for this use case, YOLOv11 performed worse than the previous versions. Below is a video that shows compares four models: YOLOv8, YOLOv9, YOLOv10, and YOLOv11 side by side (We chose the medium sized version for all models). The video shows how Vector identifies another Vector using each model. We would love to hear which model you think performed best. Please watch the video, and answer the poll below.
We, Robot
Tesla had its We, Robot event in Burbank, California last week, with a goal to showcase how it wishes to transform itself into a autonomous robots company. A short version of this event is available here. Tesla revealed its Robotaxi, Robovan, and its newest generation of the Optimus humanoid robots. Tesla CEO Elon Musk also outlined his vision of the future where Robotaxis can conveniently take us from place to place making a car a much more well utilized appliance (instead of idlying parked at someone’s home), and humanoids can do all the boring tasks such as watering plants, cleaning tables, and collecting groceries for us.
While the videos are fancy and great to watch, most media analysis believe that Tesla under-performed in the much hyped event, and that is technology behing autonomous vehicles is behind competitors like Waymo which is already available 24/7 in a few cities such as Austin, Phoenix, Los Angeles, and San Francisco. Tesla is also late in the game for humanoid robots, and its unknown how useful Tesla’s current generation of humanoid robots are.
However, Tesla makes up because of Elon’s vision, as well as having superior mechanical knowledge in building automobiles, and access to data collected by the high numbers of Tesla cars already on the streets. Irrespective, we can get a feel of how autonomous agents will change the world around us, and potentially make our lives simpler. Tesla’s We, Robots video is simply cool to watch.
Other news
The first Humanoids Summit will be held on December 11-12, 2024, at the Computer History Museum in Mountain View, CA. We will be attending the event in person and reporting back (Thanks to those to provided generous contributions and paid memberships). If you can help us by funding some of our work, please buy a paid subscription using the link below.
Petoi Inc. has comfirmed to us that that the next generation of the Nybble robot called Nybble Q will be soon available for sale and shipping after the Thanksgiving holiday. Nybble will be based of the newer generation Biboard and also feature digital servos with positional feedback. These servos can rotate according to feedback received from other servos, which makes them very useful for the robot’s legs to synchronize when doing things such as climbing stairs, or when hit with a ball. A video showing Petoi’s digital servo with positional feedback is available here.
Nybble Q will certainly make a great Christmas present. If you are interested in receiving announcements about the Nybble Q, I recommend you to sign up on Petoi’s newsletter. Petoi will also be having a booth at the Maker Faire in Vellejo, CA next week.