Inertial measurement unit

In this case, we also added a flash LiDAR for terrain classification capabilities, but we could have conceivably done that with our automotive grade LiDAR.

We used an IMU, Parker IMU GX515.

Many of you have probably used it before.

And actually, I need to credit CSIRO's Matt Hudson with the idea of using gigi cameras as the solution to USB vision.

And so what we do is we actually decide what we're going to jointly estimate the system's state's velocity over several time steps, also using an inertial measurement unit in order to correct or provide short-term spatio-temporal constraints here.

This is shown in this factor graph, wherein we connect Doppler targets to individual states, in this case, velocities of the robot.

And then we use these binary constraints in change of those velocities as the IMU factors here.

So if we want to be able to set up our full state vector, that is actually just the velocity of the robot, or in this case, just the radar, the quaternion orientation of the radar rig, and then also the biases on the accelerometer and gyroscope in order to accurately forecast or estimate what the velocity motion of the platform is.

After doing some digging, I discovered that the moment you introduce a multi-camera setup, the hardware and embedded engineers start losing hair by the handful.

I learned that there is a critical bottleneck here, which is timestamp synchronization between multiple cameras along with the IMUs or inertial measurement units soldered onto the boards.

These are essentially the six-axis gyroscopes and accelerometers found in our smartphones.

You might ask, Why make it so complicated just to record some video? Well, consider that when a worker is operating, their head and body are constantly moving.

It combines a high-performance triaxial gyroscope and a triaxial accelerometer into really a very compact, fully calibrated MEMS inertial measurement unit that's really easy to integrate in many different types of systems.

And in fact, as of the recording of this video, now the product is in production and available for customers worldwide.

What makes this product especially significant is that it delivers the kind of performance customers have traditionally associated with much larger connectorized industrial IMU systems, but in a much smaller surface mount package, which really helps reduce the kind of integration, complexity, size, weight, power, all of these system constraints that customers have.

we're seeing a big shift now especially evident towards autonomous and intelligent machines you know we've we've been investing towards this goal now for quite some time and the 16607 is just the latest imu in a multi-generational effort to support our customers as they have gone from prototypes to production whether that be in drones or robotics or industrial automation equipment, navigation systems, you know, all of these things need motion sensing that is not only accurate, but is also reliable and easy to deploy.

And so the ADI-S16607 was really designed with those needs in mind.

So this is a much more powerful, old, but still powerful board.

This one is intended to be a 9-axis IMU, so inertial measurement unit.

It not only is able to show you the acceleration in different axes, but also has a gyroscope.

it has accelerometer and it has a magnetometer so it is able to actually detect changes in magnetic field things like these are used when you need to have precise controls for example what I was thinking about using this for is to just put it somewhere on the wall fix it probably by gluing it so doing some sort of semi-permanent but hard fix and then to measure it and see what I can measure when it comes to earthshaking, earthquakes, passing cars, and so on.

So if you're out of visualization or out of line of sight for the instrument, the instrument can still track you because it tracks position through the GPS.

It has an IMU, so you don't have to basically have it in flight with the bubble, which again makes it a lot faster to capture a point.

And it also removes a major source of error that was available before, which is if you change the height of your prism, you had to key that in manually into the system.

This new system has a laser inside the pole, which is quite the engineering feature, which automatically measures the height.

Let's do a quick review of lecture five and remind ourselves where we left off.

So we started talking about IMUs or inertial sensors.

And we kind of broke it down into two main types of sensors, accelerometers and gyroscopes.

So far we've only talked about accelerometers, specifically mechanical accelerometers, and we said that from a mechanical architecture point of view, these are just damped spring mass systems in an enclosure, or a damped harmonic oscillator in an enclosure.

You need to also understand what is the state of each of its actuators.

You need to understand some of the other sensors like IMU, which is an inertia measurement unit that tells you the pose of the different parts of the robot.

And once you capture the information, first you need to synchronize all of this.

What that means is my camera is observing at this particular time.

At the same time, what was the state of my motors? What was the state of the IMUs? And you need to fuse them together.

The next thing you have to do is if you're given a task, let's assume we are running a vision language action model.

You need to tokenize this data, all of this data, the data from the current frame, but also the data from the past frames, because your next action is dependent on not just what you're currently seeing, but also a little bit of history.

Uh, you deep- you deeply covered a few areas that I often see people struggle with.

Um, your IMU chapter, which I expected to be just here's how you use an IMU, went into a lot more detail about sensor fusion and even Kalman filters.

But how is that machine learning? That's all very heuristics.

Right.

Mm-hmm

...

some other filters and how to deal with the data and how to calculate or how to use quaternions and all the things that I have done with IMU so many times.

It was kind of a joy to be able to say, "Oh, well, now I'm gonna have to...