Most of us have smartphones with apps that increasingly know where we are, and perform tasks based on our locations. Mapping apps and social media apps as us about the restaurants we frequent, or the attractions we visit. More apps than ever are offering location services indoors, even without GPS, as we walk around malls or other large indoor areas.
Could these technologies that are thriving on mobile devices be transferred to mobile robots, autonomous or otherwise? To find out, Robotics Business Review spoke with Bruce Krulwich, founder of Grizzly Analytics, about the impact indoor location positioning could make in the robotics space. Grizzly Analytics has tracked the indoor location market for more than 10 years.
Location differences in devices
Q: There are many technologies for mobile location positioning, with different accuracy levels, infrastructure requirements, and areas where they thrive. It feels like it would be easy to transfer these from a mobile phone to a mobile robot, but you feel like some of these would not work well. Why is localizing a robot different than localizing a smartphone?
Krulwich: The biggest reason is accuracy. Many location services on a phone do not need high accuracy. If you’re at the shopping mall, and an app on your phone wants to know whether you’re in the coffee shop or the shoe store, it’s fine to measure your location to within 3-4 meters. Lots of solutions based on Wi-Fi or Bluetooth signals can localize to within 3-4 meters, and do so very effectively for a phone, with low battery usage and low requirements for installing infrastructure in the shopping mall. But for robots, 3-4 meters of error is a failure.
Cleaning robots, for example, need to be able to get right up to walls and other physical objects, so they don’t have a gap of floor that doesn’t get cleaned. Warehouse robots need to be able to move down aisles and between stations without bumping into shelves and other objects in the vicinity. Home robots need to be able to move through doors without bumping into the door jambs.
The bottom line is there’s not much a moving robot can do without knowing its location more accurately than that. That’s why classic mobile technologies that have a few meters of error are not fit for use in robots.
Q: So which localization technologies from the mobile world are you seeing that can be used in mobile robots?
Krulwich: Ultra-Wideband (UWB) is making great strides in robots and other electronics. The technology got a lot of attention when Apple announced it was going to be incorporated into the next iPhone, but it has been in development for years, and is already incorporated into location-aware electronics such as robots, drones and even robotic cameras.
Q: How is UWB used for localization?
Krulwich: UWB radio signals are perfect for time and distance measurement. Most of the radio systems that we’re familiar with, such as cellular, Bluetooth and Wi-Fi, all have radio waves that go gradually up and down like a SIN wave. It is fairly hard for a radio system to measure the time or distance with waves like that, because the gradual ups and downs don’t give a very precise start and stop point to measure.
UWB radio waves are shaped like spikes, with impulse transmissions that are very precise for measuring. This means a radio system can use UWB radio to measure transmission time very precisely, resulting in a much more precise triangulation (or multilateration) of a location. UWB’s impulse radio also makes the radio signals resilient to situations of multi-path, signal reflection and other kinds of interference.
Q: In order to use these, however, measurements must be made using several different signals and beacons, correct?
Krulwich: Yes. Most UWB-based location solutions use locators, or beacons, installed around a site. These locators transmit signals that can be used to very precisely locate all UWB devices in their vicinity. This is why UWB solutions are often called “infrastructure-based”, since locators need to be installed around a site. Many UWB solutions these days are based on chips from Decawave, based in Ireland, and UWB chips are also made by NXP and other manufacturers.
Q: Are there infrastructure-free localization systems that do not require locator or beacon installation?
Krulwich: Yes – some are based on motion sensing, and some are based on the Earth’s magnetic field readings, but these are generally not precise enough for robots, similar to GPS technologies mentioned earlier.
Q: We’ve seen many robots utilizing visual simultaneous location and mapping (VSLAM) as an infrastructure-free approach. How does this fit into the equation?
Krulwich: Visual SLAM systems use one or more video cameras to literally watch the surroundings as a robot moves around, much like people see as they move around. The technology localizes itself as it moves, and also builds up a map as it moves that enables it to localize itself better the next time it is in the same area. Since VSLAM does not use radio waves, it does not need any infrastructure to be deployed, and can inherently work anywhere.
The biggest advantage of this is that many robots already have cameras, either to support interactivity, or for safety purposes. It’s also generally simple to deploy on a robot. Accuware, for example, developed their VSLAM technology originally for smartphones and mobile devices, and then deploying it in robots, drones and any other device with a camera. The results that they are achieving are very impressive.
Benefits for robots
Q: What types of robots can benefit more from these technologies than others?
Krulwich: Robots that need to localize not only themselves, but other robots, people and other vehicles moving around a site would benefit from a radio-based approach, such as UWB, which can localize many things at the same time. For example, an autonomous robot moving around a warehouse generally includes worker safety systems, where a robot will stop whenever it is at risk of hitting a person. While a visual approach can work for this by identifying people in the path of the robot, a radio approach has the advantage of being able to identify people that are not in the path of the robot, but are moving in a direction that will cross the robot’s path soon.
Of course, if a robot needs to avoid hitting people that are not wearing radio tags, or to avoid hitting people even when a radio system fails, then a visual approach is much better than a radio-based approach. Also, any robot that needs to be able to position itself in new places, or in places that for some reason can not support deployed infrastructure, will benefit hugely from infrastructure-free approaches like VSLAM.
Q: How do these localization systems compare with existing radar-like location solutions that many robots utilize, including lidar?
Krulwich: In general, the solutions I’ve mentioned are less expensive than radar-like solutions, but may not achieve the same accuracy. Laser based approaches like Lidar can achieve accuracy in the millimeter range, not in the tens of centimeters. The biggest reason for robot makers to consider these solutions is for what we call “belts and suspenders.” Multiple localization systems can often cover the deficiencies of each other, and result in overall resilience and stability.
For example, radar solutions can sometimes get confused when there are too many people, or other unknown objects in the area, which make the radar signature of a location look different than expected. Video or UWB solutions can compensate for this uncertainty very well. Because video solutions can be deployed using existing cameras and Internet connections, the overhead is very low.
Q: What impact will the recent acquisition of Decawave by Qorvo have on the UWB market? Will this have any effect on the use of UWB in robots, or is it more about UWB in other mobile devices?
Krulwich: I think Apple’s interest in UWB is all about increasing connectivity between mobile devices and other electronics that can be location-aware. This ranges from smart watches and earbuds, to robots and home control systems. A big advantage of Decawave’s (now Qorvo’s) UWB chips over Apple’s own chips is that Decawave’s are smaller and take less power, and can be more easily embedded in smaller or more power-sensitive things. Within two years, we’ll see that Apple’s UWB (and UWB in smartphones that follow Apple’s lead) will connect to a lot more robots and other location-aware things.
Q: Do you see the market for localization in robots choosing between UWB, vision and/or radar/lidar systems, or will they likely see a combination depending on the application?
Krulwich: In many cases the different technologies can be complementary, where one technology will cover for the weak spots of another. For example, lidar can sometimes be confused if there are a lot of people or other unexpected things in the area, but visual approaches can help identify the things that should be ignored, while UWB and other radio approaches are less affected (to a certain degree) by dynamic environments.
But moving forward, as robot (and drone) technologies both improve at the high end and spread into the mass market, I believe the question is cost and market. High end robots will most likely keep using the high-end technologies, with complementary technologies filling in gaps. But a huge number of lower-priced robots are coming to market that will need to localize themselves without expensive high-end technologies. Cleaning robots for homes and small businesses, for example, can install the infrastructure needed to work with UWB, along with the dead reckoning that many of them are using now. Robots in more dynamic environments, or which already have cameras for other purposes, can use VSLAM fairly cheaply and effectively, again in conjunction with dead reckoning or other low-cost approaches.
Bottom line, I think these technologies will be one of the enablers for a much wider range of low cost and more pervasive location-aware robots and other devices.