On October 1st, 2022, Tesla reveals Optimus, a robot that uses the same computer as the Tesla car. This is the first time that they show the prototype walking without any help. This prototype was developed in about 6 months and does not fully represent the concept they have for Optimus, which was also presented in its raw representation along with an explanation of what is to come in the near future.
Optimus will use the same AI algorithm that is used to train Teslas’s car autopilot, neural networks (NN). As a user of this technology since 2020, I can attest to how well it works.
The AI will be retrained for the robot purposes; i.e. to identify objects using volume rendering of the machinery a robot will have to interact with and to identify high-frequency features to estimate the trajectory of its walk. As a user of this technology since 2020, I can attest to how well it works.
The walking robot was designed with off-the-shelf actuators, but Tesla is working on their own design. The next generation Optimus was brought to the stage, and although it was not ready to walk yet, it already can make other movements. Later in the presentation, the team described the complexity of getting the robot to achieve a physical self-aware, balanced and coordinated motion, with an energy-efficient gait, and stable and balanced walk. They, however, explained how they plan to accomplish this task.
The rest of the presentation focused on explaining the differences between the robots presented and the development Tesla is undertaking to achieve its mass production goals. It was amazing to see the walking robot concept, which was developed in just six months! Now, they are working on the next generation robot rooted in the foundation of the vehicle learning process, keeping the human form, and giving it a large range of motion.
Trying to mimic all that the human body motions would be an impossible undertaking and extremely expensive. Using newly designed actuators and electrical systems, the team is working on a robot with only 28 fundamental degrees of freedom plus hand movements; this to maintain efficiency at a reduced cost for large-scale production.
The robot will have a centralized computer and power system in the torso that will minimize the downtime function and will power the bot for a full work day using a 2.3 kWh battery pack. This design will streamline manufacturing and allow for simple cooling, battery management, and safety. It also takes advantage of the current Tesla car production infrastructure and supply chain.
The brain of the robot is also in its torso, using the car’s autopilot hardware and software but changed to process visual data, making fast decisions based on sensory information and communication. It will also include hardware-level security features to protect the robot and the people around the robot. Another very important thing they are considering is damage control so the robot can fall without breaking.
A lot of the development and analysis has been devoted to the design of the actuators. These are key to allowing the robot to do regular tasks like picking up boxes or walking. The actuators were one of the main focuses of the talk’s engineering design. Taking inspiration from the human knee, the team produced a mechanical analog optimized for joint efficiency, energy, mass, and cost. They translated particular movements into joint specs, power consumption, and accumulated energy for the task vs time, and then calculated the system cost for a single actuator. Doing this for thousands of actuators allowed them to find the optimal case for each of the 28 joints. But having unique actuators per joint is not optimal for mass production, so they did a commonality study to identify the minimum number of actuators needed to fulfill the requirements for all their joints. They end up with six different actuators; three rotators, and three linear, each with large output force.
Similarly, the hand was inspired by biology, focusing on the goal to make sure that the robot can move all its fingers independently so it can operate tools. The fingers are driven by metallic tendons that are flexible and strong for power tasks and precision picking. It has 6 actuators with 11 degrees of freedom and an adaptive grasp. The hand has a controller which drives the fingers and receives sensory information to learn about the objects that are grasping and where the hand is located in space.
The goal is to develop a humanoid as soon as possible; design it with high volume production in mind, low cost, and high reliability. Elon Musk expects the robot to be cheaper than a car, less than $ 20,000 dollars.
As they still refining the robot, Musk invited people to join Tesla and help them make it a reality. The ultimate objective is to create a robot that can help millions of people by providing services and products in a safe way. Optimus should improve the economic output with benefit for the human civilization.
Full presentation
https://www.youtube.com/watch?v=Yxp1jcMn2hI