Update on Tesla's Optimus: Are we Kicking off the Race to Humanoid Robots?
Tesla's Optimus Generation 2 has truly impressive updates and progress!
Prior reading:
I’ll admit it. I’m a bit of an Elon Musk fanboy. The innovative companies he’s been a part of, and their successful track records (particularly at places like PayPal, SpaceX, Tesla, Neuralink, and X Corp) make him someone I truly find inspiring. I put him in the same category as other visionaries and innovators, like Steve Jobs, Jeff Bezos, or Mark Cuban. I even like the Cyber Truck (which is apparently controversial). So, I am admitting my bias up front. But for the sake of this article, I will do my best to check my admiration at the door and judge this project through a critical lens. This brings us to Optimus, Tesla’s ambitious Humanoid Robot project.
I first learned about Optimus during a Tesla teaser video in mid-September. I reviewed the current status of the project and it’s timeline in a previous article (for those of you who have not read this, I would strongly encourage it). A brief synopsis: Tesla announced the Optimus project back in 2021 at Tesla’s Artificial Intelligence Day event. Initially, it was not perceived seriously, but it quickly became more substantial with some seriously impressive upgrades throughout 2022 and 2023 (Figure 1). Now, as we reach the end of 2023, Tesla has announced their Optimus Generation 2 and showcased it in a captivating video (Video 1). This brief article will review some of these upgrades and assess some aspects that remain to be seen.
Video 1. Optimus Gen 2 Demonstration.
Gen 1 to Gen 2 Improvements: These improvements are either directly observed (or inferred) from the video or claimed by Tesla.
Overall Design and Functionality Improvements: The Gen 2 model is now 10 kilograms lighter, moves 30% faster, and offers many additional degrees of freedom in motion. Additionally, Tesla seems to be designing its own actuators and sensors, a significant step in customization.
Head Movement: The previous Gen 1 did not have any head movement (from what I observed). However, Gen 2 has two degrees of freedom in the neck. This innovation imparts more human-like movements to the robot.
Foot Redesign and Improved Walking: The foot design now mimics human geometry, and foot force and torque sensing have been enabled, leading to smoother and more efficient walking. The foot movement involves striking with the heel first, flattening, and then pushing off from the balls of the feet (or toes) for the next step, as evident in Figure 2. This level of articulation is part of the improvements in walking dynamics. It contrasts with the previous generation, which, as I understand, required harnessing to walk.
Improved Balance and Control: Gen 2 model can now perform a squat, demonstrating improved balance and control. This ability requires pivoting at multiple joints and managing the center of mass effectively. This is certainly a cool improvement!
Hand and Finger Upgrades for Delicate Object Manipulation: Perhaps the most notable upgrade is in the hands and fingers of the Gen 2 model. The hands are faster and now offer up to 11 degrees of freedom. The robot is also equipped with tactile feedback, enabling it to delicately manipulate objects, such as picking up and handling an egg without breaking it – a task demonstrated in Figure 3. This capability is significant as robotic grippers typically struggle with delicate and brittle objects. The heat maps on the fingers in the image suggest real-time tactile feedback processing.
Where should we be skeptical?
It's time to set aside my enthusiasm and consider some potential shortcomings. Well, I hate to admit it, but I think the current version of Optimus might have a few flaws. For starters, the robot's movements appear somewhat clumsy, and its walking and dancing skills could definitely benefit from further refinement. Additionally, its capabilities are still limited compared to what we might expect from a fully-developed AI robot. But hey, Rome wasn't built in a day, and neither will the perfect AI robot.
I also find that I am very skeptical of video-recorded demonstrations of new technology. History has shown even reputable companies like Google have been deceptive about AI demonstrations to attempt to create hype for their audience. More egregiously, the company Nikola completely misled the audience about a working car by simply rolling their car down a hill. This ‘fake it till you make it’ approach has, for better or worse, become part of the tech startup culture. Moreover, another critical point of consideration is how the current version of Optimus compares to the broader robotics market.
A criticism I have read several times is that other companies like Honda have already achieved these robotic feats (and better) over a decade ago. They argue that Tesla’s strategy leans heavily on effective marketing to their user base, akin to Apple's approach, suggesting that the features developed by Tesla are not particularly innovative but have been around for quite some time. Nonetheless, this might be where the brilliance of Musk and Tesla lies: in their ability to integrate existing technologies and create products that captivate their audience. It raises the question: if Boston Dynamics can engineer a robot capable of navigating complex obstacle courses (Video 2) but is far from market-ready or affordably commercialized, what is its practical value? To those who assert that all of this has been accomplished before, I ask: where is my robot?
Video 2. Boston Dynamics Robot - Atlas.
What comes next?
In the original article, I outlined Tesla's ambitious launch window for Optimus, targeting between 2025 and 2027. So, what can we anticipate next from this visionary project? The latest Generation 2 Optimus certainly impresses with its sleek design, complete with a polished exterior and a Tesla logo. The demonstration videos released thus far portray a robot that appears to be maturing in terms of functionality.
As we move forward, it's likely we will continue to see periodic updates through demonstration videos, perhaps on a quarterly basis. But the key milestone that many are eagerly anticipating is the deployment of Optimus on Tesla's own assembly lines. This would not only be a testament to the robot's practical capabilities but also a strong statement about its readiness for real-world applications. Seeing Optimus actively participating in the manufacturing process would be a significant leap forward, showcasing its potential impact on the industry and beyond.
References:
[1] Tesla Optimus Bot Gen 2 Gets Expert Explainer by Scott Walter - Ready to Ship!!
[2] Optimus - Gen 2
I equally love and fear to see how far these robotics have come. Coming at it from a physical therapist’s perspective though - I think people don’t realize just how hard it is to walk. People believe that the hands and upper body are the most complex, but walking is by far some of the most challenging tasks we have to teach to people. I love to see the toe push-off on Tesla’s model, but it’s almost not enough. I also see that in neither video that the “knees” never are fully extended (straight). I’m sure there’s a mechanical component as to why they’re designed to keep that joint flexed, but it makes the robots look funny. There are probably a few things they’re lacking in the models they have currently that I would say need attention to improve their gait (walking). First, they need more articulation in the feet (more movable joints) rather than one hinge joint at the ankle and the other at the toes plus a spring. Specifically they need ankle plantar flexion, the ability to point the ankle down like a dancer, to get better toe push off and knee extension. Second, the torso needs to be able to rotate forward/back as it walks. These are just two of many suggestions. Here’s a video breaking down the mechanics of gait if you’re further interested!
https://youtu.be/cu2xvCbpI3o?si=3XbaeodA7jhj70Cc
I think the competition between Elon's Team and Boston Dynamics will lead to amazing results way beyond what we see today!