From October 20 to 22, by the banks of Shanghai's Huangpu River, the 2026 Sustainable Global Leaders Conference returns once again. Nobel laureates and Turing Award winners will take the stage to discuss the future and trends of technology, and the internet-famous professors who "rescue" people from anxiety and the era of involution — Macfarlane and Xiang Biao — will hold a live dialogue!
Seats are limited, so act fast. Let us journey together toward a sustainable future! [Click to register] Author of "North America Outpost" Kang Lu, reporting from New York. After robots can run and jump, how far are they from truly entering factories and homes? On October 13, 2026, "North America Outpost" spoke with Scott Walter, Head of Robotics Research and Due Diligence at RoboStrategy, who has roughly 40 years of experience in the robotics industry, to discuss how to judge a robotics company's real capabilities and what thresholds robots still need to cross to move from demonstrations to practical applications.
Walter believes that to determine whether a robot can truly be put to use, you need to see whether it can work continuously and stably, and whether it can recover on its own after making an error. A single successful demonstration is far from enough. He points out that some robot demonstration videos use editing to skip over key operations, and identifying these missing steps is also an important part of technical due diligence.
Speaking of Chinese robotics companies such as Unitree, he says that lower-priced robot products and China's component supply chain are also helping Western researchers reduce costs and accelerate development. Walter currently leads technical research and due diligence for RoboStrategy's robotics investment projects, evaluating the hardware platforms, control systems, and engineering teams of potential portfolio companies. He co-founded Deneb Robotics in 1985; the company was acquired by Dassault Systemes in 1997 and is now part of its DELMIA business. He also co-founded Visual Components, which was acquired by KUKA in 2017.
For the Next Robot Games, I Want to See a Decathlon
Scott admits that he follows the humanoid robot games in Beijing. But when it comes to robots participating in sports competitions, is it about showcasing technology or generating attention? Scott's answer is both, "but the competition does have meaning. It provides an intuitive observation window for the outside world to see how much progress robots have made over the course of a year, and it also allows researchers to test different designs through actual competition."
However, he hopes future competitions will add tests of comprehensive capability. He says that some robots are currently designed specifically for a single event. They can run very fast, but only in a straight line and cannot turn; they can jump, but may not be able to complete other movements. These experiments are valuable, but they still do not show that a single robot can handle multiple tasks. "My suggestion is a decathlon," Scott says.
Different events such as sprints, hurdles, jumping, and throwing can simultaneously test a robot's speed, endurance, upper-body strength, and manipulation ability. It not only has to run, but also pick up objects such as a javelin and then complete a throw. A robot that can sprint 100 meters may not have the ability to finish 1,500 meters. A decathlon requires the same robot to switch between different tasks, which is closer to the capability requirements of a general-purpose robot. Humans already have a large amount of decathlon competition data, which can also provide a comparison standard. Scott hopes to use this to observe roughly what level robots have reached in terms of physical ability. But he also stresses that the decathlon is not the only option, and future events could add tasks with higher dexterity requirements.
Why Do Factories Still Need Humanoid Robots?
As for whether robots should focus on doing vertical tasks well or move toward generality as much as possible, Scott believes the answer depends on the work scenario. He says that for large-volume, repetitive tasks, specialized equipment can already do the work very quickly, and there is no need for humanoid robots to replace all of it. Homes are different. Tasks such as folding clothes, vacuuming, and cleaning are constantly changing, and none of them need to be done all day long. A robot that can do multiple household chores may be more reasonable than buying, storing, and maintaining a large set of specialized devices.
Asked why factories need humanoid robots and whether the humanoid form is suitable for factory production, Scott says there are practical reasons. Some production links are already highly automated, but people are still needed between pieces of equipment to move cardboard boxes, carry metal sheets, and prepare and replenish materials. These jobs have been difficult to automate in the past, and humanoid robots can try to take them on. "We do not want to tear down a factory that is already running and build a new one. We want to add some automation to existing factories," Scott says.
Therefore, the value of humanoid robots is not only in cooperating with people, but also in being able to take over the work reserved for humans in existing production processes. But if a factory were designed from scratch in the future, the situation might be different. Companies could rearrange equipment and material flows so that the entire process is completed by specialized automation systems, with no need for robots to look like humans. Scott uses a simple analogy. Printing once required people to set type, feed paper, and operate machines, but today it can be done inside a desktop printer. "You open the printer, and there is no one inside." In his view, some manufacturing processes in the future will also develop along this path. Humanoid robots will be an important form, but they will not become the single unified answer for all scenarios.
Figure's 56%: How Far from "Buy It and Use It at Home"?
U.S. robotics company Figure announced in September this year that robots equipped with Helix 2.5 were tested in 30 previously unseen homes on three types of tasks: tidying living rooms, folding towels, and making beds. The company said that when other test conditions were kept consistent, the full-task success rate was 56%. This figure measures the success rate of specified tasks in unfamiliar environments. Figure is also one of RoboStrategy's portfolio companies.
Asked about his understanding of the data Figure released in September, Scott says the improvement in success rate is worth noting, but robots still need to become more reliable. "In a factory, 99 might be a failing grade," because one error in every 100 operations is already too much for some high-speed production links. But errors cannot be lumped together either. If the robot is at a critical point in the production line, a single failure may affect subsequent production. If it is only preparing materials in advance, and the first attempt fails but the second retry succeeds, the situation is different.
Household tasks have similar distinctions. If a T-shirt is not folded well the first time and is folded again, users may not necessarily find that unacceptable. Therefore, to judge whether a robot is usable, you need to look not only at the success rate, but also at the consequences of failure and whether it can correct its own mistakes. For early users willing to try new things, Scott believes that an imperfect robot that can already do some things may already be attractive. He mentions that some home robot companies have already accepted reservations or begun entering households. However, he did not give a clear timetable for moving from early users to large-scale adoption. He also did not offer a specific number for how many tasks robots will be able to complete a year from now. He judges that the industry is progressing faster than outsiders can see, but it will still take time before robots are widely used. "Not this year," he says.
Doing Robot Due Diligence Means Watching Long Videos and Also Watching Failures
With more and more robot demonstrations online, how do you judge how far a company's product can actually go? Scott, who leads due diligence, shares one of his main standards: watch long, unedited records of actual operation. "I do not want to just watch a demo. I want to see actual operation," he says. He hopes companies will set up cameras, let the robot continuously complete tasks, and show the entire process. A fixed single camera angle is one approach, and multiple camera angles are also fine, but viewers need to be able to follow the complete operation process. That is because a single cut in a short video may omit the most critical step.
The robot picks up a bottle, and the next shot begins pouring something, but that does not mean it opened the cap by itself. The robot holds a blank sheet of paper and then throws a paper airplane, but that does not mean the entire folding process was completed by the robot. Scott says he once suggested that Figure release long recordings from fixed camera positions. According to his account, the company first showed a continuous one-hour logistics task video and later conducted a 200-hour livestream. He has observed that more and more companies, including China's AgiBot, have begun releasing long videos or conducting livestreams. This may also become a common way for robotics companies to demonstrate their capabilities, letting the outside world continuously observe what a robot can actually accomplish in a factory or home, rather than only watching a few minutes of successful clips.
On Unitree: Western R&D Is Also Benefiting from Chinese Robots
Regarding Unitree and the large-scale production of Chinese robotics companies, Scott believes that Chinese and U.S. companies can take different development paths. He believes that Chinese robot products are not only competing with U.S. companies, but also providing research and development tools for researchers in the United States and other Western countries. "We need these robots for research," he says. Being able to buy robots at a lower price means researchers can carry out more experiments. They can take products apart, study which designs work and which have problems, and also understand how a company solves specific technical challenges. China's supply chain can likewise help advance research. If needed actuators can be obtained within a few days, researchers can continue development more quickly.
Comments