Making a fully autonomous delivery robot (whether it flies or not) is a very hard problem. Your robot has to be prepared to operate all alone in unstructured environments, and it has to do so both reliably and efficiently. A new robot introduced this week by Piaggio Fast Forward (herein abbreviated “PFF”), a division of Italian vehicle manufacturer Piaggio, is getting in on autonomous stuff-moving, but they’re taking a slightly different approach.
Rather than try to develop a fully autonomous delivery robot from scratch, PFF is instead starting with something simpler: a pleasingly roundish robot called Gita (“gee-tah”) that will follow you around, carrying 19 kilograms of tools, groceries, or whatever you want.
Gita can carry 19 kilograms of cargo and move at up to 35 kilometers per hour, which will allow it to follow a person walking, jogging, or riding a bicycle. Photo: PFF
There are many situations where such a cargo-carrying robot would be handy. Let me give you an example: Like most people who eat food, I have to buy groceries. I live within a mile or so of a grocery store, and it would be good for my health and the environment to walk there and back. But walking places limits on what I can buy because I have to haul it back with me. Carrying a gallon of milk a mile in a bag is super annoying, and what about the rest of my groceries? So I end up driving instead. That’s where Gita would come in. The robot hauls everything, and I can walk, or jog, or bike home.
The amount of autonomy that Gita needs to follow people is limited, but not insignificant. It seems like the easiest thing to do would be to just have Gita follow a beacon that the user carries (either a phone or some kind of dedicated wearable). But PFF also wants Gita to eventually be able to navigate completely by itself, even if the user isn’t nearby (a capability that would let the robot make deliveries, like the autonomous robot delivery service being developed by London startup Starship Technologies). So the PFF engineers came up with a clever solution: Gita’s primary navigation system is a pair of stereoscopic cameras mounted on the front and back of the robot, with which it’s constantly performing visual SLAM (simultaneous localization and mapping). But rather than doing person-tracking, it’s actually comparing its SLAM picture with the SLAM that the user is doing, thanks to a stereoscopic camera on the user’s belt:
The robot is able to follow you around by comparing its own point-cloud map to another map generated by a special camera belt worn by the user. The robot and the user’s belt communicate over a dedicated Wi-Fi network. This allows approach lets the robot track you even if you turn a corner and it loses its line of sight. Photo: PFF
“Basically what the vehicles are doing is comparing the point-cloud maps that are being generated by the human operator to what the vehicle itself is seeing,” PFF CEO Jeffrey Schnapp tells IEEE Spectrum. What that means is that the robot is “leveraging the power of humans as navigators of complex environments,” he adds. “Having human operators, in essence, teach robotic vehicles [to navigate] the world, and then later setting them on relatively local autonomous missions is a smart initial way to begin to create a culture where sidewalk spaces are shared by humans and robots.”
PFF is very clear about its “human-centric” idea for Gita, which is a refreshing approach in an industry making an occasionally questionable push for full autonomy as quickly as possible. “We’re really interested in a model of vehicle design that augments and extends human capabilities, rather than starting with the ambition of replacing things that humans do well,” says Schnapp.
Gita navigates by using stereo cameras to generate point-cloud maps. After traveling a route just once while following a person, the robot might be able to collect enough point-cloud data to travel the same route again on its own. And different Gitas can all share point-cloud data with each other, so even if your particular robot hasn’t traveled a route, it will still be able to move autonomously as long as another Gita has gone before it. Image: PFF
PFF’s choice to rely solely on stereo cameras for outdoor localization in mostly unstructured environments is interesting, since there are many outdoor situations in which cameras aren’t great, like at night, or looking into low sun angles. I asked Sasha Hoffman, PFF’s chief operating officer, why they didn’t want to use lidar, which is the go-to sensor for reliable SLAM at the moment. “We think that with solid-state lidar, the price will come down, and it’ll become something that’s usable for us in the future,” she says. “Our starting price point for Gita would need to be substantially higher if we were using lidar, and we want this to be a product that’s going to be available to many more industries than would be possible if we were using lidar today.”
Gita is capable of moving at up to 35 kilometers per hour, which is fast enough to follow a moderately athletic bicyclist. Fully loaded, that represents a substantial amount of kinetic energy. “Initially, we’re operating the [robots] at pedestrian speeds, from slow walking to running,” Schnapp told us. “We want to make sure that our core usage scenario, which is very much connected to pedestrians and human motion, is rock-solid before we start tackling the challenges of following a bicycle. We’ve done a little bit of testing and we know Gitas can do this, but we’re not fully there yet.”
The robot has obstacle detection and avoidance capabilities (using vision and ultrasonic sensors), and if it gets stuck, it’ll send you a signal for help. Photo: PFF
Over the next few months, PFF will be doing lots of real-world testing of the robots in all of their modalities, which includes person-following and autonomous operation. Some of the scenarios involve assisting people who have to carry heavy things around all day in semistructured environments. Think custodians, gardeners, and other service workers in places like college campuses or retail spaces. From there, if everything goes well, Gitas will be more widely available next year, although PFF isn’t yet commenting on price. They did tell us that the goal is for Gitas to be affordable enough for consumers to acquire for themselves, and it’s also worth mentioning that Piaggio has been building complex mechanical wheely things for decades, so one of Gita’s advantages in cost is that they already have that infrastructure and experience backing them.
Without having much of an idea how Gita performs in day-to-day use, or how much it will cost, it’s too early to say whether or not it’s likely to be a commercial success. I do like PFF’s approach, though, and personally, I find the idea of a dynamic robot that can follow me and carry things to be compelling. And if you’re wondering about the camera belt thing, the company told me that the current design is an alpha version, and they expect that over the course of next year it will become much smaller—and more fashionable.