Robotic mapping

And then ended up moving to Australia to join the CSIRO, which is a national research lab in Australia and continued working on drone autonomy, eventually adding LiDAR and sometimes radar to drones for doing perception and navigation.

And then eventually, yeah, just kind of combined the use of LiDAR with SLAM on a drone and saw commercial applications that decided to co-found Innocent and spin out the technology.

And yeah, kind of gone from there eight years ago.

Okay, great.

Well, with, for example, hover map, that's more LiDAR than SLIM or it's a combination thereof.

Yeah, it's

SLAM-based LiDAR.

So we use the LiDAR data in real time.

Okay, so now compare that 28-hour marathon to the technology that's currently tempting them to just, you know, empty their bank accounts entirely.

The SLAM scanners.

Exactly.

They are looking at SLAM scanners, which stands for simultaneous localization and mapping.

Sounds expensive.

Oh, very.

Right.

And those light pulses bounce off the walls, the ceiling, the exposed pipes, whatever, and return to the sensor.

On top of the raw sensors sits the computer vision layer.

SLAM builds a map of the environment in real time while tracking the drone's position inside it.

Visual Inertial Odometry fuses camera and IMU data to hold an accurate position estimate when GPS drops out.

This is the machinery that lets a drone stay localized in a warehouse, a tunnel, or a jammed environment.

And working with NVIDIA and Linux, we built this fully autonomous humanoid.

So we're running like NVIDIA's KubiSlam, vSlam implementation, And then we added visual odometry, inertial odometry to it.

So now you can ask Ollie to go to the kitchen and bring you a coffee.

And it knows where the kitchen is and can navigate through an office space autonomously like normally you see guys hiding behind curtains with joysticks controlling their humanoid where this is fully autonomous that we were able to do with a couple of months working with NVIDIA and Linux which is pretty awesome and You know, you turn the camera, you know, full 360 and Intel's, you know, back in the robotic space.

at NexCobot.

He's got more than 15 years in industrial robotics, starting in mobile robots and SLAM research, and then moving on to PC-based robot controllers and functional safety.

So like many of us coming into functional safety as a result of our careers.

He's safety manager for the SCB100 safety controller.

So it's almost 17 years.

Yeah.

So before I start, I actually studied SLAM, the localization, simultaneously localization and mapping in my master degree.

So my research topic is actually underwater robot SLAM.

Running detection on the drone's own camera feed rather than a laptop webcam, so the operator doesn't need to be at a base station.

and fusing gesture intent with onboard autonomy so a gesture sets a goal and the drone figures out the path itself, which connects directly to the perception and navigation stack I've written about in VSLAM for autonomous UAVs and to the hands-on control work in my RC Drone Reverse Engineering series.

The takeaway, hand landmark detection is a solved, one import problem.

That's exactly why it's misleading, the tracking overlay looks like the finished product when it's really the first of three stages.

This is only really useful when you can apply it to, in the context of a moving vehicle.

If you can apply this using what's called simultaneous localization and mapping, or SLAM for short.

For those who don't know what SLAM is, basically it's an algorithm that allows you to move through an environment and map it at the same time.

So basically, at each frame, your sensor has its own local reference system, which is pretty useless for creating a map, but what it does is the SLAM algorithm takes It grabs features such as the corners of buildings, street lights, what have you.

And it can recognize that on following frames.

And so it's able to say, OK, I've moved 10 feet here.

And that's how we get those those images really easily out of that sensor.

But so as I mentioned, the sensor data can be both representative point cloud and a panoramic image.