Client
A prominent company in the smart camera technology sector, specializing in flash LiDAR, sought MulticoreWare’s expertise and support for a project involving one of their high-profile customers.
Challenges
The challenges involve taking a pre-existing perception stack (implemented in ROS) utilizing the customer’s ToF camera and:
- Porting the stack from an x86 development environment to an Nvidia Jetson AGX ORIN ARM architecture.
- Enhancing frame rates (FPS) by optimizing code to leverage GPU acceleration.
- Implementing a custom data logger application to comply with US regulatory standards for safety systems.
- Integrating an Over-The-Air (OTA) application utilizing 5G cellular connectivity for reliable in-vehicle data upload/download.
Solution
X86 to ARM Porting
- Reviewed the existing software components developed for the x86 platform.
- Assessed the compatibility of the existing software components with the Nvidia platform, considering differences in architecture, instruction set, and hardware capabilities.
- Determined & implemented the modifications and adjustments needed to port the software components. This included code optimization and adaptation to the ARM architecture.
Performance Optimization
- Profiled the stack and identified the key areas where GPU utilization would be beneficial.
- Leveraged our experience and expertise with GPU offloading and CUDA code to implement these functions on the GPU.
- Profiled specific high-load functions with and without GPU offloading to determine the most efficient approach for the expected data.
Datalogger and OTA Application
- Researched and selected the most appropriate hardware solution (5G cellular module) for the requirement and project constraints.
- Developed a secure and robust data logging system from scratch, designed to handle edge cases, which uploads all system data to the cloud using the 5G cellular module in case of a system alert.
Business Impact
The collaboration with MulticoreWare yielded significant outcomes for our client.
- Enabled them to demonstrate a fully functional system to potential end-customers on a real vehicle platform.
- End-to-end perception stack operational on ARM devices.
- Accelerating the total FPS of their solution made it applicable to many industries.
- Utilization of GPU for accelerated inference, resulting in a remarkable 25% performance boost.
- Implementation of a resilient data logging system with edge case protection ensures data integrity during critical events.
Conclusion
In conclusion, MulticoreWare demonstrated exceptional proficiency in GPU and high-performance perception software optimization. Our team successfully ported and optimized an end-to-end perception stack and developed an independent data logging system. This project highlights our expertise in perception stacks across various platforms and our capability to create reliable software for any environment.
Discover how we can help you achieve innovative results in Autonomous Mobility and Robotics. Contact our team at info@multicorewareinc.com
