California announced that come April, it will be legal to operate robocars with no safety driver inside. There will need to be a remote operator, monitoring the vehicle at all times during testing.

That's a far cry from a short while ago when California proposed banning such cars, which is part of what encouraged Google to move to Arizona for testing. And while most people are thinking about what this means for Waymo and the other more advanced companies, it has another big consequence in delivery robots.

As I have written before, I'm involved with Starship Technologies delivery robots which run on the sidewalks rather than the roads. Our robots are very small, slow and light, which means they can't hurt you even if everything went wrong and they hit you. Road robots are big enough to hurt you, but will go faster as long as traffic is not heavy.

Since Starship announced, there have been a huge number of companies appearing to attempt delivery. Most have also been sidewalk based, but two recent entries have shown plans to go on the road:

• Nuro, founded by my friends Jiajun Zhu and Dave Ferguson who I worked with at Google, has shown an impressive half-width robot that goes at slow-street speeds and has several delivery compartments.
• Udelv has a tiny delivery van which currently can hold a safety driver. They plan to sell this vans.
• Some other stealth companies are also working in this space.

On the sidewalk, many companies are trying all sorts of sizes. Companies like Marble have a fairly large and heavy robot. Robby (Palo Alto) has a robot very similar to the Starship one, particularly their 2nd generation one. Only a few are trying small and light robots, most notably Kiwi of Berkeley. Most are going for higher end vehicles, with expensive LIDARs.

The Starship and 2nd generation Robby use a 6 wheel design that can climb curbs. Most others use 4 wheel or even "2 wheel plus 2 casters" designs which will be more limited to flat surfaces.

A couple of robots are going for a two wheel design, balancing like the Segway. This two wheeled design may be able to also go over small bumps. This includes the TwinswHeel from France and the Yape from Italy.

Segway itself has a design which takes their new "Loomo" scooter (which includes autonomous features) and bolts it to a cargo box, making it no longer a 2-wheel balancing act.

And as I wrote earlier, at least two companies plan to build full produce stores on wheels that come to the customer so the customer can pick out their items. A more real one has been announced in Santa Clara.

It's impressive how many different approaches there are. That's because it's not hard for a small team to put together a basic small wheeled robot and give it some autonomous motion capability. That's what everybody in every robotics club has been doing for a decade, though mostly indoors. A series of contests called the Magellan prizes have been rewarding this for a decade, though for a long time results were highly disappointing. Robotics is reaching the level where this is easy to do.

There is, unfortunately a lot more to do to make autonomous out in the real world reliable and safe, and teams are busily working on it.

At the same time, cities are reacting in different ways. Many places have overtly legalized the robots on their sidewalks, but San Francisco decided to ban them. You should of course be wary of what I say about a competitor, but I fear that Marble, which was the first company to test in San Francisco, has a larger and heavier robot that is more likely to cause concern.

Having the ability to test without a safety driver makes robots the size of the Nuro possible. It is not designed to have a human inside it, so it can only be tested this way. So far they have had to do all testing of the real robot off-road. (They test the main system on-road by putting the sensor package on regular cars.)

In the future, we'll see competition between street robots, sidewalk robots and of course, flying drones. Drones will have lower cargo capacity but good speed. They also have more trouble landing. They will probably focus on the market for light, urgent deliveries. The public may not tolerate a sky full of drones overhead for all deliveries.

Sidewalk robots will be most useful in the shorter distances. They are the most energy efficient and lowest cost, especially when doing a solo delivery for one customer. One of the great advantages of robotic delivery is the idea that the robot comes on your schedule, not that of a driver. That's much harder to do if you want to load a robot with packages for several customers at the same time. Since the robots can't yet leave packages at the destination (or even get to the door in most cases) the customer must be home and come out to the robot.

The on-road robots will also have trouble in dense areas which don't have driveways or parking spaces in which to stop.

In some cases, road-based vehicles and sidewalk robots (or even walking ones) could combine. For example, a van or truck full of robots could stop in an area, release them, and then another truck coming by later could pick up the empty robots for return. That truck might have a human driver and still be somewhat efficient. This was a concept behind a pilot project Starship and Mercedes did for "vans & robots."