A humanoid at the end of year 1 is not the same robot it was on day 1. Every robot will age differently. Friction changes. Backlash grows. Sensors drift. Structures loosen. Batteries age. Impacts accumulate. Parts are maintained or replaced at different times. So how should a robot keep its understanding of its changing body up to date? Condition monitoring? Predictive maintenance? Recalibration? Online system identification? Digital twins? The hard question is not only: can the robot do the task? It is: are we sure it still knows what its body can do?

"Think about your client's client." @humane_robot

Russ Tedrake, CEO of Walden Robotics, opened The Economist’s Physical AI & Robotics Summit with a "How did we get there?" intro.

First panel of the Physical AI & Robotics Summit, at @EconomistEvents. Key outputs: - use cases should drive form factors. - black factories might be the future, but not the present. - next gen actuators are needed to overcome the limits of existing solutions.

The body is part of the message. Humans constantly read motion. Someone slowing down before reaching us communicates something different from someone approaching at full speed. Direction, posture, speed and distance all help us anticipate intent. Robots will need this too. A good trajectory may not only be efficient and safe for the robot. It may also need to be understandable to the person standing next to it. Motion is not only execution. It is communication.

General intelligence. General body? We often assume that a general-purpose robot should also have a general-purpose body. But bodies are always compromises. A greyhound and a mastiff are both dogs, but their bodies are optimized for very different things. The same will be true for robots. Reach. Payload. Speed. Compliance. Hand design. Energy use. Stability. As robot intelligence becomes more general, will robot bodies also converge toward one general form? Or will we see increasingly specialized bodies driven by increasingly general intelligence? General intelligence may not require a general body.

Will humanoid robotics stay full-stack? Early quantum companies often had to build almost everything themselves. As the field matured, specialists appeared: Qubits, Control, Error correction, Algorithms, Applications. Humanoid robotics may follow a similar path. Today, hardware, control, perception and AI are still tightly coupled, so full-stack makes sense. But if interfaces mature, specialization becomes possible. Full-stack may be the fastest way to build a product in an emerging market. It may not be the most scalable way to build an industry.

Skin or structure? A humanoid’s exterior does not have to be just a cover. It can sense contact. Add compliance. Spread impact. Manage friction. Protect the mechanisms underneath. As robots work closer to people, the outer surface becomes part of the interaction system. The question is no longer only: What should cover the robot? But: What should the cover do?

Rigid torso in humanoid robotics ? Keep your feet planted. Now reach behind you. Your torso joins the movement. A rigid humanoid torso simplifies design and load transfer. An articulated one can extend reach and redistribute motion, but adds mass and control complexity. How much mobility is worth putting between the arms and legs?

A big ❤️ for the clip. A big 👎 for the question starting the post. This is a bad marketing move from whoever is behind BD's X account. There might not be one humanoid robotics race, as several use cases will create several markets.

Why should one actuator serve only one joint? Robot actuation is often organized joint by joint. Biology often couples them, like with the hamstrings spanning hip and knee. These muscles can transfer force and energy across the limb instead of treating each joint independently. Independent actuation makes robots easier to design, model and control. But as humanoids mature, will we keep that modularity, or use more multi-joint actuation?

We know how to price used cars because we understand mileage, service history, accidents and wear. What is the equivalent for a humanoid? Operating hours? Number of falls? Payload history? Joint replacements? Thermal history? Calibration drift? A second-hand robot market will eventually need more than a maintenance log. It will need evidence of what the body can still do.

…some rideable quadruped robots already borrow the round barrel of a horse. A horse has a wide ribcage because it has lungs. A robot does not. That is form over function again. And the rider pays the price. A horse-like body makes the riding position more demanding at the hips and knees. A motorbike keeps the legs much closer to hip width. So if rideable quadruped robots become a thing, their body design may need to look less like a horse and more like a motorbike with legs, like Corleo, the concept from Kawasaki. The machine may be robotic. The rider is still human.