Welcome to the Automotive Benchmark Task Force – a joint effort between AVCC and MLCommons.

AVCC and MLCommons have issued an open call for participation – industry players, we want you to be at the forefront of automotive vehicle technology benchmarking. Join us!

We are here to define and develop an industry-standard ML benchmark suite for automotive to be used in requests for information (RFIs) and requests for quotation (RFQs). OEMs and automotive suppliers regularly send out RFIs and RFQs to vendors to understand a solution’s compute performance and system resource utilization. This is a crucial step when assessing whether the sourced part is a suitable choice when designing the next-generation automotive compute platforms.

Why is AI/ML important for automotive designs?

  1. Safety – AI/ML enhances safety features such as collision avoidance, lane departure warning and other intelligent driving features. Measuring AI performance ensures automotive systems to meet specific latency and other performance KPIs.
  2. Efficiency – AI/ML is used to optimize vehicle performance, including fuel efficiency and predictive maintenance.
  3. User Experience – Consumers have a better experience in a vehicle with smart function.
  4. Regulatory Compliance – Automotive regulation is very strict and collaboration is key to meet regulations around the use of AI/ML in vehicles, particularly in relation to safety. Industry driven alliance keep us ahead of regulatory requirements.

Why are MLCommons and AVCC Partnering on Automotive AI/ML Performance?

  • AVCC has a proven track record of publishing Technical Reports on how to measure ML performance for automotive
  • MLCommons has a proven track record of developing well adopted industry standard ML benchmark suites Edge, Datacenter and Mobile
  • Inaccurate AI/ML performance measurements increase financial, technical and safety risk
  • The Automotive Benchmark Task Force (ABTF) is focused on defining the right AI/ML performance targets and specifications to ensure fairness, accuracy and broad industry adoption

Why do we exist?

The benefit of having an industry ML automotive benchmark suite is two-fold: it makes it easier to do fair (so called apples-to-apples) comparisons between different technologies, and help guide where to focus engineering efforts for developing future hardware and optimizing current software. These benchmarks will help drive the industry forwards faster.

Technical Approach

Measuring automotive AI/ML performance requires focus to specific areas in the image processing chain. The following figure illustrates how the ABTF is focused strictly on PERCEPTION processing for the first benchmark demonstration.

 

The ABTF benchmark will leverage the existing MLCommons LoadGen infrastructure running on dedicated AI/ML automotive hardware. The following diagram illustrates the high-level LoadGen control concept.

Deliverables

The WG will create a MLPerf benchmark suite for automotive running on systems developed for automotive purposes. The focus will be on, but not limited to, camera sensor perception as this is currently the most mature type of ML for automotive. However, the use of ML in new areas is increasing as more features adopt more sophisticated algorithms. With time, the benchmark suite will be updated to take these new use-cases in consideration.

Meeting Schedule

Weekly on Wednesday from 8:05-9:00AM Pacific.

Task Force Chairs

Kasper Mecklenburg – LinkedIn

Kasper Mecklenburg, based in San Jose, is a performance analysis engineer at Arm within automotive with focus on software-defined vehicles and on enabling AD/ADAS applications. Kasper has a Bachelor’s in Numerical Analysis and a Master’s in Applied Physics from Lund Institute of Technology, Sweden.

James Goel – LinkedIn

James Goel leads the AI/ML standards group at Qualcomm where he is a Senior Director of Technical Standards. He is on the board of the Video Electronics Standards Association (VESA) and is the chair of the MIPI Alliance Technical Steering Group. Previously, he was the VP of Engineering at Silicon Optix before it was acquired by Qualcomm in 2011. He holds a Bachelor’s of Applied Science (B.A.Sc) in Electrical Engineering from the University of Waterloo in Canada and is a licensed Professional Engineer (P.Eng.) in Ontario where he’s practiced for the last 31 years.

For More Information

If you need additional information, please reach out to us at [email protected] to learn more!