This week we look at a new way to create 3D maps of the world from satellites. We examine an alternative to the current GPS system and we find out how drones are learning to navigate in new environments without human assistance. Finally we look at the amazing development of Large Language Models such as ChatGPT. How quickly are they actually progressing?
3D Maps of Earth
Florida based startup, Nuview is developing a 3D map of the Earth’s surface. Currently satellites provide a 2D view of the planet. Nuview will shortly launch a constellation of satellites that will maintain a 3D map of the ever changing surface of the Earth.
The LiDAR based system will offer much more detail that previously available. Using pulsed lasers the system measures the time that it takes for the laser to be reflected by objects on the surface. It is able to send those lasers trough trees and other types of vegetation to gain an accurate mapping of the surface.
Currently only 5% of the Earth’s surface has been mapped with LiDAR. A project to measure ice-sheet elevation in the Arctic was launched in 2018. Advances in space technology including the rapidly falling cost of launching satellites has allowed this project to become feasible. A constellation of 20 satellites will collect data 100 times faster than current commercial aerial solutions.
The data can be used by farmers to optimize crop yields and water usage and urban planners to create better environments in cities. Nuview claims to already have US$1.2million in signed contracts for data from a range of different types of customers.
Alternative to GPS
The current GPS system is owned and operated by the US Military. The military can turn it off at any time for any reason suited to their own purposes. This provides a risk to the many commercial uses of GPS that have flourished since the military opened GPS data feeds up to commercial providers.
A team at Ohio State University have developed an alternative system that uses the many satellites in Low Earth Orbit. Each of those satellites emits a variety of signals that can be listened to without being able to read the signal.
The team found that by listening to the signals for 10 minutes their algorithm was able to identify the receiver on the ground within 5.8 meters. The researchers did not need permission or assistance from the satellite operators to use the signals. The team only used publicly available information related to the satellites downlink transmission frequency and a rough estimate of the satellites location.
So many of our vital systems now rely on GPS. This includes communications systems, the power grid, emergency services, navigation systems and food delivery services. GPS signals are weak and can be easily spoofed. This provides a security risk in safety critical situations. Self driving cars and other automated systems will only amplify the limitations of the current system.
Tens of thousands of satellites will be launched into Low Earth Orbit in the coming years. These mega constellations will revolutionize numerous technologies and benefit a range of applications. The current algorithm is also able to identify where the satellite is in space without the satellite revealing its’ location.
The current GPS system is very mature. As this new system improves the technical capabilities of the algorithm it will begin to provide a viable alternative particularly for mission critical applications for when the US Military wants their system back.
Autonomous Drones
There are many different drone displays that have become commonplace in big celebrations (e.g. New Year Celebrations, the recent British Coronation). However these displays are preprogrammed and the drones only fly exactly where they are told to fly. A team at MIT have now used Liquid Neural Networks to allow drones to navigate themselves in the wild.
We have spoken previously about Liquid Neural Networks (LNN) and how they enable robots to travel into areas that have not previously been traversed. Robots, drones and other automated devices are usually trained on the area in which they work. This limits their ability to change environments.
LNN’s were inspired by organic brains. In an organic brain when it is exposed to new data, the synapses strengthen and thus improve the network. LNN’s contain a set of artificial neurons and synapses that can continuously adapt to new inputs. The team used the Caernorhabdtis elegans worm for inspiration, it has just 302 neurons and 8,000 synaptic connections. A much smaller brain than the average human however it is enough for the drone to navigate in an unknown environment.
The team were able to make the drone pass a number of tests including fly to target, stress tests, target rotation and dynamic target tracking. The LNN was first taught to identify a red chair, leading the drone to recognize and fly toward the chair. The drone was able to do this from up to 45 meters away.
The LNN performed better than 3 other types of neural networks that were tested. The brain inspired LNN showed less drifting from the target than other more traditional neural networks.
The team’s goal is for the system to be able to identify other objects, including humans. Disaster response would have a massive increase in effectiveness if drones could quickly arrive at a disaster site, recognize humans in trouble and provide immediate appropriate aid (e.g. flotation devices, emergency medical supplies etc.).
How quickly is AI progressing?
We are currently experiencing a flood of publicity around new AI applications but is this a new set of capabilities or just publicity for a few new toys. Firstly a little history.
I was at a meetup a few months ago where one of the main presenters said in their presentation “AI has been around for years, even as far back as 2014 people were talking about AI”. Actually it was in the 1920’s that humans first proposed that machines could be taught to think. Alan Turing (of Enigma fame) invented his Turing machine in 1937. It was at a conference at Dartmouth University in the US in 1956 that the term Artificial Intelligence was first coined.
So is the technology behind the latest AI sensation, ChatGPT new or old? The following chart shows the progression of the 5 major components of Large Language Models (LLM’s, ChatGPT is an example of a LLM but not the only one on the market).
Progression in new capabilities tends to follow a similar path, they start out very slowly off a zero base however they eventually find their way and then progress at an astounding pace. We are currently witnessing the astounding pace part of the LLM development curve.
I want to be clear here, LLMs can only repeat what they have been trained on. They are very good at that, I use ChatGPT every day to do just that, it is much faster (and in many ways better) than a Google search, which essentially does the same thing. LLMs however can not create new thought. They are a long way from sentient, they do not currently think for themselves. Currently!
Paying it Forward
If you have a start-up or know of a start-up that has a product ready for market please let me know. I would be happy to have a look and feature the startup in this newsletter. Also if any startups need introductions please get in touch and I will help where I can.
If you have any questions or comments please comment below.
I would also appreciate it if you could forward this newsletter to anyone that you think might be interested.
Till next week.
Also the other military can jam GPS. This sounds much harder to jam,with many different frequencies.