The Humanoid Robot Games events are a market map

Most of the sporting events require full autonomy. Some don't. The Games organizers are saying: bipedal running is solved, hurdles & obstacles are still hard, and explosive movements are still hard. Most of the work scenarios can be done either autonomously or via teleoperation ... but you get double points for autonomy.

Here are the sporting events

In the 100 meters, a human at the controls is disqualifying. But in the 100-meter obstacles, it isn't. That's a major facet of robot development revealed in one line of a schedule. 

Twenty of the 30 competition events now require full autonomy. That includes all six flat track events: 100m and 400m in both size classes, the 1500, and the 4×100 relay. It also includes all three soccer formats, the gymnastics floor event, street dance, both dancesport events, both wushu events, both tug-of-war classes, both pitch-pot classes, and table tennis.

Notice what unites the mandated locomotion events: none of them require a huge amount of exteroception in the control loop. Flat-track running is a proprioceptive problem. A policy trained in sim with domain randomization, running blind on IMU and joint encoders, will get a biped down a known lane. Organizers also cut the 100m time limit from three minutes to one, the 400 from 15 to five, and the 1500 from 40 to 15, which converts "finished" from last year into "finished at a pace that implies a real gait policy rather than a shuffle" this year.

The autonomy-mandated non-locomotion events are all closed-world: a choreographed routine, a rope, a pot, a fixed table.

10 events with no autonomy designation

Ten events carry no autonomy designation: both obstacle races, all three jumps, both weightlifting classes, both kickboxing classes, and the comprehensive pentathlon.

These fail for three different reasons.

The obstacle races are a perception problem. Put a hurdle on the track and you need exteroception fused into a dynamic controller at speed: terrain estimation, foot placement replanning, all inside a gait cycle. This is the actual research frontier.

The jumps are a hardware problem. Standing long jump, long jump and standing high jump aren't blocked by perception; they're blocked by actuator power density and landing impact. Nobody is protecting a research question here ... they're protecting expensive hardware from a bad landing.

Kickboxing is an adversary problem. A non-stationary opponent optimizing against you is the one thing sim-to-real transfer handles worst, and it comes with a safety envelope nobody wants to be responsible for. Weightlifting is similar: maximal torque, thermal limits, zero margin for a controller error.

Here are the work scenario events

All 20 work scenario events use a multiplier: autonomous ×1.0, remote-controlled ×0.5.

This is better mechanism design than a ban, and it's honest about where the industry actually is. A 2:1 price keeps every team in, forces each one to publicly declare which mode it's running, and makes the autonomous-vs-teleop delta legible across the whole cohort.

It also quietly preserves the teleop pipeline. 

Half-credit runs still happen, and teleoperated demonstrations in real hotel rooms, real factory cells and real retail floors are precisely the input that imitation-learning and VLA pipelines consume.

The scenario list itself is a market map

We see events for:

• retail service
• F&B service
• assembly and feeding
• packing and warehousing
• emergency response
• fire rescue
• office general service
• household service
• hotel guest service
• library book sorting
• garden management
• EV charging

This is pretty much a deployment-order bet. NEV charging is the pretty big tell here: it's a somewhat structured environment, presents a repeatable insertion task, and it's import to where China is already going.

Eight of 20 scenario events are hands

Power tool assembly, powder weighing, bricklaying and block stacking, nail fastening, cap and lid opening, box and package unpacking, tweezers bean picking, cable connection.

That's the largest discipline block in the Games. It's bigger than track and field, and that makes perfect sense. Cable connection and nail fastening are line work; powder weighing is lab work; tweezers bean picking is a fine-manipulation benchmark with a long history.

The distribution is the message. Legs are solved to the point of being a spectator formality. Value is in hands and fingers.

What still isn't measured

No stairs, no ladders, no unstructured or outdoor terrain.

There's no endurance or battery-life event, no human–robot physical handoff, no standalone human-giving-verbal-instruction-and-robot-follows-orders test. And there are no metrics anywhere for cost, energy per task, reliability across a shift, or MTBF.

In other words, there's room to grow for next year.

This year, the schedule went from 26 events to 51; teams from 280 to 666; robots from roughly 500 to 2,056. Soccer moved from what organizers called "preschooler" to "youth" ... dribbling on the move, diving saves. 

What about next year?

Expect more mandatory autonomy and more challenging work tasks.

The obstacle races are the ones to watch. Expect more difficulty. On the work side, watch the multiplier. If the teleop discount gets steeper next year — or disappears entirely for the easier scenarios like book sorting and retail service — that's the organizers saying autonomy has arrived for those tasks. 

And watch for the metrics that don't exist yet: energy use, battery swaps, repeated success.