Welcome to the 6th installment of our Introduction to Autonomous Mowing series. We will take a shallow dive into the sensors we use this week and next week. The aim is to help you understand why we have chosen the specific sensors we have integrated and the levels of redundancy these sensors offer. First, we will dig into the GNSS (GPS) sensors this week. Next week we will look at hopefully more than one of the other sensors – IMU, resolvers, LiDAR, stereo cameras, ultrasonics, and cell communications onboard Nomad.
We have discussed this before, but the nerds that we are love acronyms and unusual terms. Quite often, we gloss over exactly what these terms are. This blog will use a bunch of these terms, but we think (hope) we have expanded their definitions enough that it will all make sense.
We discussed this in general terms a few posts back – but we felt we needed to expand on what GNSS or a global navigation satellite system is. GNSS is the term we use today to describe not just the U.S. GPS (Global Positioning System) but also the GLONASS (Russia), Galileo (European Union), BeiDou (China), QZSS (Japan), and IRNSS/NavIC (India) systems. The Japanese and Indian systems provide local country coverage. On NOMAD, we have several GNSS receivers that only use the four primary global constellations (or collections of satellites) – GPS, GLONASS, Galileo, and BeiDou.
Note that any GNSS receiver has an antenna and usually a separate printed circuit board (PCB) receiver. These systems receive data from the satellites within the various constellations. The receivers primarily receive a signal containing orbital information about the satellite vehicle (SV) that transmitted the signal and the time the signal was sent. If we know the time it took to travel from the satellite (SV) to our antenna on the mower, and we know the speed of transmission through space – then we have a distance. If we have several of these ranges from SVs somewhere in the sky (some of us are old enough to remember when we were pleased to see 5 SVs on an early GPS receiver), we can then start to determine our position. When we talk about GNSS constellations and determining position, we also talk about the geometry of where these SVs are with respect to each other. Good geometry is where the SVs are spread out across the sky. Bad geometry is if all SVs “in view” are off to one side of the mower – the dilution of precision (DOP) is a term you may hear when people talk about this. We also spend a lot of time focusing on the degradation of the accuracy of these systems. One of the most significant components we look at is the variations in our estimate of the “speed” with which the signal traveled from the SV to our antenna. These SVs are orbiting between 25,000km and 30,000km above the earth. The signals they transmit travel through multiple layers of varying “space.” We have to understand the variability in the speed of transmission through the ionosphere, the troposphere, the clock errors in the satellites, the clock errors in our receivers, any errors within the estimated position given for the SV, etc.
One way to deal with all of these precise details is to use some form of corrections that answer all of the “what about” issues raised above. A simple point-to-point set of corrections often used with GNSS is real-time kinematic (RTK) corrections. In simple terms, you set a GNSS receiver up and a very well-known location (on a tripod or hanging on a fence at a known point) and send the differences from the known truth that this static receiver determines to a mobile GNSS receiver (say on our mower) that is being impacted by the same variations in the transmission paths and this receiver removes those errors. This is how we can end up with a real-time positioning accuracy around a few centimeters or an inch or so on the mower, even with all of the unknowns mentioned above. So we could (and do) position our mower using a locally delivered set of corrections from an RTK base station. The industry uses the term "base station" for the fixed location and "rover" for the moving location.
These corrections (Radio Technical Commission for Maritime Services [RTCM] corrections) can be sent over a simple local radio link from a single base station. We argue amongst ourselves about this – but a rover working much more than 20km or 12 miles away from the base station will start to see errors in the position accuracy as the signals are traveling through a very slightly different path from the SV’s, and the corrections start to degrade.
In most built-up areas around the world, commercial companies have set up networks of these base stations that can be subscribed to, and these networks will provide a known level of accuracy through their coverage area, as well as (important for our application) a known level of redundancy and reliability in the provision of these services. This data is often delivered in very small packets (lengths of a message) over a cell phone link. This means we must have a cell phone (or cell modem) onboard the rover (NOMAD) to receive these corrections.
These corrections are usually provided in an NTRIP (Networked Transport of RTCM via Internet Protocol) format. Most GNSS receivers today are pre-configured to accept these types of corrections.
What do we do outside of good cell coverage or an area where these RTK corrections are readily available? We have been working offshore in the oil and gas business for the past 15 years and have used differential or GNSS corrections out in the middle of the ocean that was delivered over other satellites. The communications satellites (as used for your Dish TV service or satellite phone services) are also used like cell networks to send these corrections. For the early systems, we had to have two receivers with two antennas to download the GPS signals and the second system to download the corrections. Today the industry has moved on, and we can download corrections over the L1 frequency band on the same antenna as we are using to receive the GNSS data. These wide-area services are known as precise point positioning solutions or PPP. L1-delivered PPP services commonly deliver positioning accuracy of 5cm. BUT – (always a but) The difference between the various services offered varies significantly in the time the positioning solution takes to “converge” or reach this accuracy. Why do we care about convergence time?
What does a mower spend a lot of time doing (apart from mowing grass)? They spend much time moving in and out from under trees or next to buildings or tall fences. All these things, trees especially, block out the signals from SVs, and the GNSS solution disappears or fails. If we drive back into the open where we see the satellites, we want the position solution accuracy to converge (or reconverge) very quickly – within seconds. If it takes 2 to 3 minutes (not unheard of) for a PPP solution to converge, we will probably never recover precise GNSS for much of the day as we are constantly moving into and out of some form of tree canopy over the mower. Some RTK solutions can also take a long time to converge. On Nomad, we have a primary GNSS system that works with RTK or L1 PPP corrections. This system has a speedy convergence time.
We also have a dual antenna GNSS solution on Nomad that provides a backup to our primary GNSS system and our “heading” (which direction we are facing). These two antennas are connected to GNSS receivers at the PCB level to determine an accurate heading for the mower. We need accurate heading as we must take the information measured from our various sensors (LiDAR and Cameras) bolted to the top of the mower and translate or move this data via software into the absolute coordinate reference frame the GNSS systems work in. Any error in heading turns into an error in the translated position of this other sensor data – which, as far as the user is concerned – relates to less production as we have to allow for more mowing deck overlap to ensure we have cut the grass.
We hope this did not make you snore too much, but as we say – the devil is in the detail. We look forward to your comments on our Introduction to Autonomous Mowing series. Look out for upcoming topics, including Stories from the Field.