A Compilation of Robots Falling Down at the DARPA Robotics Challenge

I remember thinking that the humanoid robotics revolution was imminent after seeing ASIMO walk around 2010[0], but after observing the slapstick comedy that was the DARPA Robotics Challenge around 2015[1], it seemed that we had actually made zero progress. But I haven't felt that way in years! My feeling at this point is that the best humanoid robots have, from a hardware perspective , surpassed the "minimally viable product" threshold. By this I mean that they are physically capable of completing enough tasks that would make them worthwhile a worthwhile product. They are physically capable of walking the dog, folding the laundry, picking up the living room, bringing in the groceries, etc. It's possible that there are reliability issues to address, but given that the newest generation of bots are fully electric, it's hard to imagine that this is a more difficult problem than those they've already solved. At this point, we're just waiting on the software to catch up. And surely this problem seems far more doable today than it did 20 years ago, given all the progress we've made in machine vision and machine learning! We'll know that the robotics revolution is truly imminent when the big players start pouring capital into development. Tesla has already joined the fray. I expect more will follow in the next few years. [0] link [1] ▶ this video

It was always software. Humanoid robots could walk up and down stairs and use basic tools, in preprogrammed lab conditions and/or by manual teleoperations since 90s[1], and in theory autonomously since 2010s[2]. Nowadays they just can, thanks to NVIDIA CUDA, but grabbing objects is still sketchy[3]. 1: link 2: ▶ this video 3: link

The progress since the 2015 DARPA challenge (where robots succeeded mostly in falling down) is impressive. Less than a decade! ▶ this video link

We've come a long way from the 2015 DARPA Challenge, where the robots succeeded only in falling down: ▶ this video link

That reminds me of Moravec's paradox: What we thought was hard and difficult (and indicative of intelligence), such as playing Chess and computing derivatives, is actually quite easy, while a lot of things that we consider trivially easy (distinguish a dog from a cat, pick up an egg, read some squiggly text) are quite hard. link One of my favourite demonstration of that was the 2015 DARPA robot challenge, where one robot after the other failed at such difficult tasks as walking and opening a door. ▶ this video

> I'd trust a machine to drive more accurately than most people I see on the highway. Humans aren't special, in fact more often than not we're sloppy, subject to fatigue, and a whole bunch of other negative things. Humans are much, much, much more capable than the absolute state-of-the-art robots when it comes to doing things in an uncontrolled environment. ▶ this video

Moravec's paradox - stuff we thought is hard (chess, derivatives) is easy, while stuff we think is easy (face recognition, getting out of a car, opening a door) is really hard. link One of my favourite videos is this compilation of robots falling (failing) during the simplest tasks at the 2015 DARPA challenge: ▶ this video

slightly off-topic, but the images in the article illustrate it beautifully: I think many people don't realise (I didn't) what an amazing skill it is to be able to read handwriting (cursive). I took it for granted to be able to read hand-written German, English, Spanish, and totally forgot that it must have taken me some time as a kid to acquire that skill (beyond reading and writing generally). However, when I studied Chinese for a bit, the (printed) characters are tricky to start with, but reading a fluent writer's handwriting ("unconstrained", as they call it in the article) is nigh impossible. I think stroke order is hugely important, in that it forms the conventions for fast abbreviated writing. For example, the "mouth" radical, a square, "口", when written quickly basically looks like a lower-case "b", because the left lower stroke comes first, then top and right in one stroke, then the bottom stroke. More complicated characters basically become a squiggle, and it's really magical, if you think about it, that you can just read it. I mean, look at figures 5, 6, 7 in that article. For me, a nice illustration of Moravec's paradox of AI [1]: what we think is difficult (chess, differentiation, ...) is easy, and what we think is easy (reading handwriting, putting an egg in the fridge, walking (!) and opening a door [2]) is hard... [1] link [2] ▶ this video ("A Compilation of Robots Falling Down at the [2015] DARPA Robotics Challenge")

Yup, there is a boatload of low level knowledge missing from any AI that we build. It is painfully visible with robots (and autonomous cars are robots as well), see DARPA challenge video which was then lead BTW by Gill Pratt: ▶ this video A lot of stuff we take for granted such a supporting oneself by a nearby wall when loosing balance, is completely out of reach of todays "AI".

This [1] list of failures shows that some robots seem to "just" have hardcoded motions, for example for turning the valve. [1] ▶ this video