Intelligent drone simulation programme takes learning even deeper

Autonomous drones and UAVs are designed to make decisions without input from a pilot or operator. They don’t adhere to a single algorithm, instead they make intelligent, real-time decisions in complex and changing environments to determine the best possible outcome.

Companies use the programme to give drones a particular scenario, or scenarios, based on their own unique need. The simulator then runs through thousands of different possibilities to determine the best outcome.

Aviation experts, Bell, wanted to use the simulator to train their UAVs to land autonomously. To do this, they needed to run two types of simulation. One to determine how the drone should land, and another to help the drone to determine where it is safe to land. Once this is determined, the AI will then work together to complete the process.

Something which could look like solid ground when a drone is in flight could prove to be an unsuitable and unstable landing space once the drone descends, such as a field of corn. For a drone to try to land there, it would prove hazardous, time consuming, and costly. Therefore, training a drone in a simulator is a safer, faster, and more effective way to carry out these tests.

The saved time, low cost, and safety of using a drone in a simulated scenario before real-world deployment is a huge draw for companies.

Another benefit for companies is the programme can make employees daily work easier. By using the simulation programme, companies can gather the data they need to train their AI, meaning that people do not need to be experts to be able to use, operate or understand the technology. They may not be data analysts, for example, so they can give all their real-world information and experiences over to Microsoft and using their AI, it finds the best solution based on the information it receives.

The programme works by using machine learning, deep reinforcement learning and simulation, and offers companies a safe space to be able to test, track and train how their drones will behave in real world scenarios without having to cost vast expense or man hours.

The simulator has thousands of environments and billions of parameters that companies can use to create an autonomous flying machine before taking the expensive technology out into the real world.

The programme needs to know: Where am I? Where am I going? And, what do I do next? It then runs through thousands of scenarios in a minute.

Some of the potential dangers or hazards it will look out for are:

• Weather conditions
• Landing areas
• Static or moving objects

Gurdeep Pall, Microsoft Corporate Vice President for Business Incubations in Technology & Research said: “Autonomous systems will transform many industries and enable many aerial scenarios, from the last-mile delivery of goods in congested cities to the inspection of downed power lines from 1,000 miles away.

“But first we must safely train these systems in a realistic, virtualised world. Project AirSim is a critical tool that lets us bridge the world of bits and the world of atoms, and it shows the power of the industrial metaverse – the virtual worlds where businesses will build, test and hone solutions and then bring them into the real world.”