# General-purpose telerobotics needs work that’s not suitable for a for-profit company * Key arguments * General-purpose robotics is hard and takes a long time. [[Boston Dynamics]] built [[Big Dog Robot]] in 2005 with funding from [[DARPA]]. They revealed [[Spot Robot]] in 2016 and started selling it to the public in 2020. The time and money scale to get Spot from prototype to production makes sense for a for-profit organization. However, it’s unlikely that they would have been able to build spot without a big chunk of the DARPA-funded work in previous years. Also note that as soon as Boston Dynamics figured out the right design regime for a legged robot, you have knockoffs quickly coming in. * Because [[There are five heavily coupled areas of improvement for telerobotics technically]], the work that needs to be done is this “design-space exploration” work that Boston Dynamics was doing pre-spot. At the core the questions are “do you use simulations or automation as a crutch against latency?” “How much do you do each and how much do you work on minimizing absolute latency?” Different answers to these questions and the resulting design decisions put you in very different areas of design space. * More generally, general purpose robotics plays tend to be terrible investments or pivot to be more specialized. * [[Boston Dynamics]] * [[Willow Garage attempted to do crowdsourced telerobotics]] * [[Rethink Robotics]] * [[Fetch Robotics]] * [[Kindred AI]] * Arguably these companies would have had more impact if they had been more open. * The tightly coupled nature of design choices in telerobotic subsystems means that the project needs the flexibility to design and build its own hardware for each component. This was the case for early computing as well. All the current approaches either use off-the-shelf parts ([[Converge Robotics Group]], [[Sanctuary AI]], all the [[Avatar X-Prize]] teams) or built their own hardware for a demo ([[Toyota T-HR3]] and [[Centauro Project]]). As a result, I doubt any of them will make meaningful steps towards general-purpose telerobotics. * Specialized robots can make good companies but despite broad claims to be working towards full generalization, you’ve never seen them make the jump from specialized to general. * See * [[Cobalt Robotics]] * [[Hello Robot]] * [[Intuitive Surgical - Da Vinci Robots]] * [[Fugro]] * [[Kraft Telerobotics]] * [[Haddington Dynamics]] * [[Kindred AI]] * A PARPA program would enable research work/coordination across more specialized organizations with serious contexts of use like Oil/Gas ROVs on the pieces of the technology that are not on their critical path without having to go in and compete with them. Instead of being yet-another-startup in the competition grind, a PARPA program could play a unique role among companies working on telerobotics adjacent things by acting like the second level of a multi-level hierarchy. Companies are fairly blind optimization functions and a PARPA could act as a forcing function that nudges them out of local optima and towards a global optima. Kind of like a dropout layer in machine learning. * The organizational culture you need to do research engineering is very different than the one you need to specialize and sell a product into a high-uncertainty market (as opposed to pharmaceuticals or space, which have very little market uncertainty). * A PARPA program does not preclude spinning off a company once we’ve actually figured out the right area of design space. Of course there is no bullet-proof argument that a project should not be organized as a for-profit company unless there is zero probability that it will produce anything that looks like a product. There is a non-zero chance that you can get lucky and hit on the correct solution early and a market quickly. However, there is evidence that going straight for a startup has a lot of downsides and a PARPA program has several unique potential upsides. The work to Is there an immediate and clear market for a general-purpose telerobotic system? The work to do general purpose telerobotics is going to require a lot of dicking around with hardware Perhaps the closest analogy is [[Boston Dynamics]] — the work that they’ve done is amazing but once they got to something commercial (spot) ie they figured out the right design regime, there are almost immediately knockoffs. From the standpoint of the world, this is great — dog robots everywhere. From a business standpoint it would have been much better to start a company once they figured out the spot thing. It would be much better from a world standpoint to be less secretive about how the atlas technology works. People need the resources and space to experiment with different design regimes and build hardware with serious contexts of use but without the Once you figure out the right design regime As far as I can tell, none of the companies have actually *figured it out* Focusing on demo videos General purpose telerobotics is not going to happen through one weird trick A for-profit company would need to work closely If you want to put millions of dollars in and not expect a return No niche is valuable More reality tests Nobody has actually said how they would do anything differently from the [DARPA robotics challenge](https://www.darpa.mil/program/darpa-robotics-challenge) [YouTube](https://www.youtube.com/watch?v=g0TaYhjpOfo) https://www.cs.cmu.edu/~cga/drc/jfr-what.pdf ### Related * Probably a duplicate with [[A telerobotics startup is not a good idea]] * [[When should an idea that smells like research be a startup?]] <!-- #evergreen --> [Web URL for this note](http://notes.benjaminreinhardt.com/General-purpose+telerobotics+needs+work+that’s+not+suitable+for+a+for-profit+company) [Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/General-purpose+telerobotics+needs+work+that’s+not+suitable+for+a+for-profit+company)