Challenge
As vehicle systems become increasingly complex and interdependent, the required control strategies need to take in huge amounts of vehicle data. For overall vehicle control algorithms, the dynamic vehicle model needs a high level of accuracy, while also being able to deliver simulation results in real-time speed. Engineers at Potential Motors needed a high-fidelity, multidomain vehicle model that could effectively train their new AI-based control systems for electric vehicles.
Solution
Engineers at Potential Motors chose MapleSim for the creation of their vehicle dynamics models. By choosing MapleSim, they could combine several vehicle domains in a single, unified model that would help train their AI-based control system. These models could be easily exported as executable C-code, a key requirement for their team’s AI-training workflow.
Result
Potential Motors was able to successfully incorporate a high-fidelity vehicle model into their controls training workflow. By training their AI algorithms against the MapleSim model, they could run extremely fast simulations to validate and optimize their control strategies. The MapleSim model now forms a central part of their test platform, allowing for faster, more optimized control systems for electric vehicles.
For over a century, automotive manufactures have invested huge amounts of resources into virtually every detail of a typical vehicle. As the vehicle itself has developed into a high-tech marvel of engineering, so too has the corresponding software that now controls almost every aspect of how the vehicle itself operates. In these modern vehicles, advanced algorithms monitor for performance issues, alert you to safety concerns, and are beginning to operate without any human input whatsoever. For these advanced control systems, however, engineers need to ensure that every aspect of vehicle performance can be factored into the overall control system.
Specializing in electric vehicles, Potential Motors develops vehicle control software that brings all the control data together and forms a unified, intelligent connection between the driver and vehicle. Their RallyAI vehicle control software works with the latest electric control and sensing hardware to operate in parallel with the driver for a safer, more reliable driving experience. The control software uses the latest artificial intelligence (AI) and machine learning (ML) techniques to continuously train and improve their system.
Potential Motors develops AI-based vehicle software that brings all the control data together and forms a unified, intelligent connection between the driver and vehicle.
To develop their control software, engineers at Potential Motors needed to train their AI-based algorithms against models that would simulate the performance of electric vehicles. These models need to incorporate the dynamics of multiple vehicle subsystems – the powertrain, suspension, batteries, and more. By testing against this vehicle “digital twin,” the overall control software can be significantly optimized before physical vehicle testing needs to begin.
As CTO and co-founder of Potential Motors, Isaac Barkhouse knew that a dynamic vehicle model would be central to the development of their control software. “When dealing with electric vehicles, you’ve got the typical automotive domains (electrical, thermal, mechanical, software, and so on). Being able to simulate and understand all of those domains together, in a unified way – to build a digital twin of the system – allows us to understand the vehicle before testing against the real thing,” he noted when describing the development of RallyAI.
Isaac and his team chose MapleSim, the multidomain modeling and simulation tool from Maplesoft, as the tool for building their dynamic vehicle model. Using MapleSim, they could create high-fidelity, physics-based models of various vehicle subsystems, and use the unified model as a platform for their algorithm training. The MapleSim model was created with a combination of drag-and-drop physical components, and customized components created to suit their specific needs for control testing.
By using a dynamic vehicle model created in MapleSim, the team at Potential Motors could perform faster simulations and more quickly iterate on their control software development.
To effectively train their AI-based control algorithms, the engineers need to run a massive amount of simulations that provide valuable data for the algorithm. For that reason, the simulation speed of their vehicle model would be crucial. In addition to providing an intuitive modeling environment, MapleSim was also chosen for its ability to generate highly efficient simulation code. By automatically performing a variety of simplification techniques to the model, MapleSim generated simulation code that was exported in executable C-code. The exported C-code was then incorporated into their training algorithms, effectively providing the control software with the virtual vehicle upon which to simulate performance.
Being able to simulate and understand these domains together, in a unified way - to build a digital twin of the system – allows us to understand the vehicle before testing against the real thing,
Isaac Barkhouse, CTO, Potential Motors
By training their AI-based control software against digital twins from MapleSim, Potential Motors is well underway with their goal to redefine the driving experience for electric vehicles. While the company is still relatively new compared to many in the automotive sector, their technology has already attracted serious interest from the major players. They’ve also recently received $2.5M in funding from investors who share their excitement for this technology. Potential Motors is continuing to develop their technology with their quickly growing team and unveil initial demos to the public in 2021.
Contact Maplesoft to learn how MapleSim can help with your projects.
Products
Industry/Application Area
