Are you building a robot that’s supposed to autonomously navigate in a useful way? Cool, that means you’ll be needing a LIDAR system, then. For better or worse, it’s usually just that straightforward: LIDAR is arguably the best sensor we have right now for reliable navigation, localization, and obstacle avoidance for ground robots. In terms of relatively low-cost sensors, sonar is poor resolution and short range; structured light and time-of-flight sensors are short range and don’t work well outdoors; and camera-based vision systems aren’t robust enough for reliable navigation.
The “relatively low cost” bit is the problem: LIDARs are pricey, and an “affordable” 2D unit, with a range of 10 meters or less, can cost you over US $1,000. This is an enormous problem for both hobbyists and cost-conscious commercial robotics developers (i.e. every single commercial robotics developer).
A San Leandro, Calif.-based startup called Scanse has developed a 2D LIDAR system that promises to be simultaneosly much cheaper and much better than what’s out there. For $250, you get a spinning LIDAR sensor with a range of 40 meters, even outdoors. We featured Sweep on Video Friday a while back, but we wanted follow up with a more detailed article, with some additional info from the Scanse team.
LIDARs are being used in almost every commercial or industrial mobile robot in operation or under development today. Once you start looking for them, you see them everywhere, from simple material-handling robots to advanced autonomous cars. LIDARs are also used extensively in applications besides navigation and localization: you can use them to make detailed 3D scans of areas, for example, or to establish safety curtains. Whatever the application, a LIDAR is often the single most expensive component in any robotic system, and that’s where Scanse saw a huge opportunity:
What makes Sweep unique is that, at its core, it relies on a new kind of LIDAR sensor, developed by a company called PulsedLight. Scanse co-founder Tyson Messori explains how it works:
Sweep is based on a new time of flight ranging method, which involves sending out laser pulses that are made up of a series of micro pulses. These micro pulses act as a kind of light based checksum, which allows the sensor to more easily correlate returning light to the known pattern, and achieve a phase difference measurement. This method is much different than traditional LIDAR, which uses a train of identical pulses to measure the phase difference between outgoing and incoming light.
The traditional method requires the pulses to be much brighter than ambient light, otherwise it gets drowned in noise. This requires high power lasers and finely tuned detectors. The new method we are using allows the sensor to use lower power components, which contributes to its low cost. This also gives it the ability to sense surfaces up to 40 meters away, even in noisy sunlit environments. A final added benefit comes from the fact that each pulse packet is unique, which means the sensor can reject multi-bounce returns (as they would come out of order), as well as light from adjacent sensors.
The disadvantage of this technique is that Sweep spends more time on each individual measurement, which lowers its update rate to 500 Hz. Similar low-cost LIDAR systems typically offer update rates of 2,000 Hz to 10,000 Hz. What does this mean for you? Here’s what Messori says about use cases:
Sweep’s lower update rate makes it most suitable for obstacle detection rather than environment scanning, if used on moving vehicles. A typical mobile robot use case will set Sweep to scan at around 3 Hz, giving about 2 degree angular resolution. Static scanning use cases don’t care as much about scan speed. An example would be a surveyor looking to make 3D scans of environments. Paired with a second rotation stage, Sweep can provide a full sphere of data points with 0.75 degree angular resolution in about 3 minutes.
To get a sense of how Sweep stacks up against other low-cost LIDARs, here’s a table that compares Sweep’s specs with the SICK TiM561, the Hokuyo URG-04LX, and the RPLIDAR, which looks to be a re-engineered version of the Revo LDS from inside a Neato robot vacuum:
A detailed spec sheet on Sweep can be found here.
Messori notes that Sweep also includes hardware and software that make it easy to add brains and cleverness to your robot:
One of the key innovations unique to Sweep, which differentiates it from all the sensors listed above, is its ability to perform much of the processor intensive data filtering internally. Sweep has an ST Cortex M line microprocessor dedicated to angle/range data fusion and range filtering and another smaller microprocessor dedicated to motor control. This allows it to offload much of the data filtering from the parent controller.
Many consumer robots still have fairly low power processors, which can get bogged down with the amount of data LIDAR sensors can produce. By performing some of the filtering internally, Sweep can be used on much smaller and lower power platforms. One of the first filters we are providing is the ability to ask it to only output where the closest obstacles are. In this way, an aerial drone or other small robot can easily avoid obstacles with minimal processing effort. Scanse will be releasing additional filters targeted toward the surveillance and IOT fields.
This is one of the primary benefits of Sweep. While you could, if you wanted to, buy the LIDAR sensor made by PulsedLight that’s inside Sweep (called LIDAR-Lite), getting Sweep as an integrated package of hardware and software has some big advantages. Scanse says it has invested thousands of hours into their platform to make it compact and rugged. Perhaps most important, they’ve also done a lot of the software work for you: Sweep will play nice with ROS, will come with an SDK for Arduino and Raspberry Pi (and other languages), and drivers for Pixhawk, NI’s roboRIO, and more.
We should note that PulsedLight was acquired by Garmin earlier this year, but Messori isn’t worried about it: “We have known the sensor development team that is now part of Garmin for more than two years and have maintained a strong relationship with them,” he says. “The sensor component going into the units we will be delivering in Q4 2016 are a future version of the technology, which have been upgraded with benefit of Garmin’s resources.”
As of this morning, Sweep has over 770 backers, and is at about $200,000 in pledges, just shy of their $230,000 goal. With only four days to go on this (the campaign ends on Monday), it would be great to see this get funded—there’s just a desperate need for it in robotics. That said, as with all crowdfunded projects, it’s important to understand that Sweep is not a ready-to-ship product: you’re not buying one, you’re pledging money to Scanse to help them create and deliver a product in the future. While we have a generally favorable impression of this thing, there’s always a risk that for whatever reason Scanse won’t be able to make it happen. If you believe in what they’re trying to do, committing now will help push them over their Kickstarter goal and put you in line for the first batch of Sweep units. Your other option is to wait until next year, and (hopefully) buy one off-the-shelf for a little extra instead.