Covariant.ai and applying deep learning to robotics

Here are two example sentences that catch my attention:<p>1. &quot;Most robots these days make use of some form of Deep Learning.&quot; This is not obvious. What is the basis for it?<p>2. &quot;Robots themselves have been around forever, but, with a few exceptions, have been disappointments.&quot; In historical context, this is hardly true. Look at assembly lines, at automation, just to start.

Just for some context from someone who is involved in robotics, both Google X and Samsung Research have research teams working on robotics arms. I would expect to see a lot more of these companies in the coming years, weaving a narrative of RL ( currently getting hyped a lot in academia, again ) and factory automation.<p>Manipulation is another task that appear deceptively simple, but is actually very complex for machines, similar to autonomous driving. Personally, any solution involving manipulation with <i>fingers</i> cannot be viable. Thankfully their approach appear to use a simple gripper. Most of their publication is around general RL (<a href="https:&#x2F;&#x2F;covariant.ai&#x2F;our-approach" rel="nofollow">https:&#x2F;&#x2F;covariant.ai&#x2F;our-approach</a>). And again similar to AVs, the sim to real gap is pretty big here too.<p>One good thing is that warehouses is a more constrained environment and can be further structured around specific robots. And Amazon has internal robotics teams and have deployed robotic arms in limited settings. It works there because the entire warehouse is structured around robots, that&#x27;s what it takes.

Another impressive startup in this area is nomagic.ai. From what I know, they are more advanced than covariant, had been in production for more than a year and recently raised a decent Seed round.<p>Good luck to both teams!

Meta&#x2F;off-topic: I think that for many people the use all these cool-sounding &quot;mathy&quot; names - covariant, differential, tensor (flow), etc. may in fact be more irritating and confusing than justified to any meaningful degree.

From my research, maybe the most interesting characteristic is that all these companies seem dependent on the sucker gripper. I haven&#x27;t worked in this field, but you&#x27;d think it&#x27;d be easy getting other gripper to work well, especially since covariant is combining simulated training with non-simulated training.

I was under the impression that reinforcement learning already tackled the &quot;picking&quot; problem sufficiently well.

&gt; <i>the technology was shockingly advanced ... we were blown away</i><p>This is a very impressive step on the road, but this kind of hyperbole always sets off my <i>Segway early-warning-system</i>.