Jonathan Roberts
In the past few days, your social media feeds may have been full of videos showing humanoid robots sprinting, jumping and performing various impressive (and occasionally clumsy) feats. This is because the second World Humanoid Robot Games have just concluded in Beijing.
While the games featured many spectacular sights, some of the more mundane events offered deeper insight into how this technology will eventually assist humans.
Commonly referred to as the robot Olympics, this event began last year when about 500 robots showed off their abilities across 25 events. This year’s edition featured more than 2,000 robots competing in 50 events. Under the rules, competing entries must be humanoid: possessing a torso, two arms, two legs, an onboard power source and no trailing cables.
Half of the events are direct copies of the human Olympics and Paralympics: track and field, weightlifting, football and boxing. Just as in the human games, the crowd-pleaser was the 100m sprint. This year, two robots beat Usain Bolt’s world record of 9.58 seconds, with the fastest clocking 9.39 seconds.
A similar narrative played out in the 400m, where one robot finished in 39.7 seconds — well clear of the current men’s world record of 43.03 seconds and the women’s record of 47.60 seconds.
Impressive as these athletic feats are, the remaining 25 non-sporting events paint a far more compelling picture. This list is particularly telling about how humanoid robotics is developing, or how manufacturers hope it will evolve.
The non-sporting lineup includes book sorting, garden management, bricklaying and perhaps the most intriguing event: tweezer bean picking. These skills are based on practical, real-world scenarios outside the sporting arena, which is what makes these games so interesting.
For a humanoid robot to reliably complete a complex task, it must first see the objects it needs to interact with. This alone is a major hurdle; lighting conditions that do not trouble humans remain difficult for robots.
If a robot sees the object it needs to handle, it must then work out how to pick it up, which involves judging both large and small forces. Again, this is very challenging. It then needs to figure out where and how to move the object, place it down and plan its next move. Executing this sequence accurately every time is tough, as each step is difficult in its own right.
The hardest part, however, is error recovery. A robot must recognise when something has not gone according to plan and then figure out what to do next.
It is therefore the non-sporting events at the Humanoid Robot Games that serve as the true test of how close humanoids are to becoming useful everyday tools. This year’s games exposed precisely these sorts of difficulties: a cable resting at the wrong angle, an object slightly out of reach, or a package shifting unexpectedly. Robots must learn to spot these real-world imperfections and recover without a human stepping in.
The Conversation