Robotics – Worse, Slower & More Expensive

Mr Wang gets over his skis.

• The illusion of dancing robots seen at every trade show and at the robot Olympics is leading people to think that robotics is much closer to commerciality than it really is, and to a sceptic, Mr Wang’s comments look aimed at supporting the share price rather than a realistic prediction.
• Mr XingXing Wang (Founder, CEO Unitree Robotics) is predicting that the breakthrough in robotics is going to come in 2028 despite having stated in August that it was going to take 10 years or more.
• The breakthrough he is referring to is a generalised robotics model that would allow humanoid robots to reliably and safely navigate the human world without having to be trained for every SKU and every upgrade.
• This is the holy grail of robotics, as when the problem is solved, domestic house bots will become a commercial reality, unleashing what I suspect will be a market much larger than the automotive market.
• However, solving this problem is proving to be notoriously difficult, and so something substantial needs to have happened in order to bring that forecast in by 8 years or so.
• According to Mr Wang, it is the sudden leap in performance of frontier models that has changed his mind, but this is not something that is really borne out by the data.
• Looking at the benchmarks, the latest models from OpenAI and Anthropic are somewhat better than their predecessors but not materially so.
• Furthermore, their ability to recognise when they are making things up has also not materially changed from previous versions, although there has been some improvement.
• One can also see this from looking at the robots sent to almost every trade show, which have not materially changed since I first saw them over 1 year ago.
• Somewhere around every demo pen lurks an operator with an adapted gamepad and a list of combo key actions (like Street Fighter) that instruct the robot to perform actions like shaking hands, walking, doing a forward roll and so on.
• Hence, these robots give an impression of autonomy, but in reality they are little more than remote-controlled toys.
• The real robots that have some degree of autonomy are the ones that are stacking shelves, making coffee or handing out gifts, and all of these are slow, stilted and constantly make mistakes.
• This is the real state of where robotics is today, which leads me to conclude that Mr Wang would have been better served by holding onto his previous opinion, although it would probably make his shareholders less happy.
• Hence, I see no reason to change my long-term view, and my tests for the advent of robotics also do not need to change.
• This test states that when robots are better, faster and cheaper than humans, they will rapidly replace humans in the tasks where robots can pass all three criteria.
• So far, the robots are worse, slower and much more expensive, meaning that they still have a long way to go.
• The company I keep the closest eye on at the moment is Agility Robotics, as it sells robots to companies today and therefore represents a reasonable idea of where the industry finds itself.
• Robotics offers a great opportunity, but I have no position in it for no other reason than it is still so far away that it will be other things that influence the valuation of companies that may be exposed rather than the sudden emergence of a new industry.
• Robots are on the horizon, but they are going to take a very long time to arrive in the present.