Our simulation environment is built with an equivalence module at its core. Equivalence is a technique for measuring the discrepancy between a simulated environment and its real world equivalent by using performance measurements on an algorithm that is going to operate on data sensed in that 3D environment.

Trust in the outputs is paramount in order to have confidence in simulating critical business decisions and validating safety parameters.

Radar is a pivotal sensor to unlock reliable and cost-effective full stack autonomy. Combining this with lidar and camera provides you a full suite of potential autonomous sensors.

These establish performance more easily before committing to build or evaluate potential providers and reduce real world data collection costs.

It is nearly impossible to recreate all the possible scenarios required to validate autonomous solutions in the real-world.

Generating scenarios enables running thousands of simulated scenes as part of millions of specific scenarios, allowing fast virtual design iteration often before the physical build.

Example 1: Warehouse automation

The importance of this simulation was to understand the trade offs between 3 potential ways of loading planes using automation. We use real physics implementations to measure KPIs like throughput, efficiency, and shock.

The outputs of the simulations can then be used in early business justifications for large-scale capital decisions - in this case, how many planes to order.

Example 2: Flow analysis for smart roads

In this example we simulated overall traffic flow as we transition from human drivers to autonomous vehicles.

We found autonomous driving reduces congestion as expected. However, if car manufacturers work together on setting parameters such as risk tolerance, global congestion can be reduced by a further 1-3%.

Example 3: Algorithm performance verification

In this example, we're measuring the performance of a Simultaneous Location And Mapping (SLAM) algorithm that uses stereo camera sensors to build up an internal map of the vehicle's environment, in different environmental conditions.

In this case, the algorithm struggled to meet 'all of the spec all of the time' - the simulation accelerates development of a more robust algorithm and shows where mitigations are required to meet the safety case.

With modular architectures, we can be adaptive for specific use cases and requirements. This allows for flexibility by swapping relevant modules which will increase interactions so that the ecosystem can work together towards common system performance goals.

Simulation Architecture

Our simulation architecture is built upon fundamental building blocks that can be utilized and adapted to suit your specific needs. You can even swap in or out modules as required, reducing redundancy while enabling smooth integration with larger ecosystems.

What happens when people need to be “in the loop autoplay” of the simulated test environment? Using VR we can test this not only for safety benefits such as AI assurance, but also to cost effectively reach a wider test audience.

Example 1 - Human reaction testing

Here we created dangerous level 3 autonomy driving scenarios in the simulator which require handover of control from an AD algorithm to human to determine whether their response is worse compared to them having been driving the whole time.

Example 2 - Dynamic navigation testing

Here we tested how a robot can adapt to changing scenarios. VR can be used as a tool to get a more realistic simulation of the human in the environment.

Example 3 - User experience testing

Here we validated the overall proposition from a wide set of users. This allowed us to iterate different designs and prioritise features without the need for physical prototyping, complex logistics in a rapid and geographically scalable way.