The Humanoid Robot Games Events Are A Market Map

When the second World Humanoid Robot Games opens at Beijing’s National Speed Skating Oval on August 22, the headline numbers will be about scale. The Games will have a record 2,056 robots, up 311% year-over-year, and 666 teams, up 138%. Sixteen countries and regions from six continents will participate in 1,301 individual matches over five days, though 96% of the entries are Chinese

That’s impressive, but it’s also mostly beside the point.

The schedule of events itself is the big news, not who competes or who wins. It’s one of the most honest documents in robotics right now, because it reveals which capabilities are solved and can be done by robots autonomously … and which ones are harder and still need a human at the controls, AKA teleoperation. And it’s a market map of what jobs China expects humanoids to be able to accomplish.

Here is the full program, and then what it reveals.

The games are split into two halves: competition events modeled on human sport – like the real Olympics – and scenario-based events modeled on human work. Out of the 50 published events 30 are competitive sports-like events, and 20 are work or job-scenario-based. Many of the competitive sports events require full autonomy, while for the scenario-based events, Games officials will reward autonomy with full points score while punishing teleoperation by cutting reward points in half.

The competitive sporting events

Here are the competitive events:

The 11 track and field events are mixed: basic running races require autonomy, while the obstacle and jumping events have no such mandate. Soccer and gymnastics require autonomy, as do the dancing events, boxing and Tai Chi events, and table tennis. Weightlifting, kickboxing and the comprehensive pentathlon do not require autonomy, according to the schedule.

What the organizers of the games are telling us is that flat-ground bipedal running is now a commodity: competitors must do it hands-off. They’re also saying that hurdles, jumps, maximal-load weightlifting and kickboxing are harder and can have a human in the loop. In other words, closed-loop perception at speed, explosive dynamics with a hard landing and adversarial contact are not yet solved.

The working scenario events

Here are the scenario events:

The work scenario list is very different. Here, competitors get full points by completing events autonomously, but only half points for teleoperation.

What the organizers of the Games are telling us here is that retail, food service, assembly, warehousing, hospitality work and household are where they expect humanoid robots to get deployed commercially first. This is essentially a market map in game form.

It’s especially interesting to look at the dexterous hand challenges, with eight events. Here, robots will be competing in events like power tool assembly, powder weighing, bricklaying and block stacking, nail fastening, cap and lid opening, box and package unpacking, tweezers bean picking and cable connection.

All of those hand events look super-interesting for rewarding robots with great hands and top-notch physical AI to drive them. None of these are sports: they are jobs, or at least key tasks in jobs.

But, note what is missing – and therefore probably too challenging for most humanoids to do yet – no stairs or ladders, no unstructured terrain, no endurance or battery-life event, no human-robot physical collaboration, and no metric anywhere for cost, energy per task, reliability across a shift, or mean time between failures. There’s also no standalone test of instruction-following: giving a robot verbal orders and seeing if it can complete the required task.

In other words: you’ve come a long way, baby (err, humanoid robots). But there’s a really, really, really long way to go yet.

Growth this year is in the work scenarios

Year over year, the event list almost doubled, 280 teams became 666, and roughly 500 robots became 2,056. The 100-meter time limit fell by two-thirds. Soccer went from what organizers described as “preschooler” to what they think will be “youth” level, including dribbling on the move and maybe even some diving saves.

But almost all of the event growth is in the work half, not the sport half.

As Wang Peng of the Beijing Academy of Social Sciences put it: “Last year it was about whether they could run to the finish line. This year it’s about whether they can get the job done.”