TRED Simulation, Environment, Data and API

For Dstl and specifically the Defence AI Centre Experimentation Hub (DAIC-X) delivery team for concept 1 there is a requirement to showcase and demonstrate AI-services that control a platform with stealthy behaviour as well as having the ability to react accordingly to a dynamic scenario to enable future exploitation of the AI-services involved by the [REDACTED] project.

This behaviour is to be referred to as being "threat aware".

The first step is to utalise a simulated environment that will allow for ideas to be developed and experimended with in a safe and cost effective manner.

A previous [REDACTED] simulated environment, developed by [REDACTED] , was used to act as a hardware-in-the-loop demonstration where real-time performance of solutions could be tested and proved for an un-crewed air vehicle (UAV) before proceeding to trials.

This existing simulation environment is a suitable starting point to be able to generate AI-enabled "threat aware" behaviour, but there is a requirement to add to and adjust its current capabilities in order to be effective.

With the understanding that development of this dynamic "threat aware" behaviour has already begun the requirements for this work comes in two stages which allow for access to the initial environment, terrain data and Python API in stage 1 and an updated/refined versions in stage 2.

The requirement for each element have been detailed below.

Terrain Data: To allow for an efficient and iterative approach to developing the concept, the delivery team require accurate and representative data to be provided/accessible that will act as a conduit to deploying an AI-service trained in early stages to be able to perform well in the full simulation environment (for further details see The Unity Environment section).

The data will be required to contain environment terrain information to include at least the height and type of the terrain, natural elements and man-made structures.

This will be able to be used in conjunction with the delivery team's own risk-maps in order to train AI algorithms to be able to complete a mission whilst being "threat aware".

Python API: The Python API that is designed to be used in the final delivery of the simulation should be provided so that development of the AI-services can progress in the direction of the future simulation software, allowing for an easier transition between Stage 1 and Stage 2 deliverables as the environment gets updated.

The API should include config and a post mission report formats.

This will allow for the delivery team to configure suitable experiments and provide information for further analysis on the missions, timestamped location data should be accessible for all of the entities within the environment.

The Unity Environment: 1/ Realistic Terrain, 2/ Physics and Control and 3/ User Configurable Entities.