The University of Massachusetts Amherst


Associate Professor Tongping Liu of the UMass Amherst Electrical and Computer Engineering Department and his two collaborators – Liu’s Ph.D. student Jin Zhou and Dexin Lee of Inceptio Technology – have won the Best Paper Award at the 2023 Institute of Electrical and Electronics Engineers International Symposium on Autonomous Vehicle Software in Tokyo, Japan. Their winning paper, titled “Profile Dynamic Memory Allocation in Autonomous-driving Software,” describes a groundbreaking software tool called “MemTrace,” which addresses the technical challenges for software designers on how to leverage existing software from artificial-intelligence research and autonomous-driving development and make it useful, reliable, and efficient.

“The landscape of autonomous driving has been evolving quickly in the past decade, from lab prototyping to mass production, and from close-road testing to commercial services on open roads,” as Liu and his associates explain. “In this process, the software takes a central role in defining the autonomous-driving capability and ensuring the safety of the passengers and the [surrounding] people.”

One key question for software designers is how to implement autonomous-driving software efficiently while maintaining high confidence in the functional safety of the vehicle.

As Liu and his colleagues explain, software is critical in autonomous-driving systems. But, currently, there is a deficiency in research that “quantitatively assesses the impact of dynamic-memory management and evaluates the trade-off between the flexibility provided by dynamic-memory management and the safety of static-memory allocation in autonomous-driving software.” Both dynamic memory and static memory have their advantages and disadvantages for the software.

A key question for the designers of autonomous-driving software is how to combine dynamic-memory management and static-memory allocation in a safe, effective, and reliable way. As Liu and his colleagues observe, there is a critical need for a tool to effectively profile the memory allocations within autonomous-driving software and understand their effects.

Liu’s team says that MemTrace is the first technology that addresses the challenge of estimating the trade-off between dynamic-memory management and static-memory allocation in autonomous-driving software to help designers make informed decisions about dynamic-memory allocation.

According to Liu and his colleagues, “This paper presents a software tool called MemTrace to conveniently analyze the dynamic-memory-management behavior of autonomous-driving software and provide important analytical results for software designers to make judgments on software quality, run-time efficiency, and safety with high confidence.”

The researchers add that “Our experimental results show that MemTrace can effectively provide detailed per-iteration results for autonomous-driving software modules and identify potential memory-related hazards.”

As Liu’s research team concludes, “Combining all these analytic results, software designers can have a deep understanding of dynamic-memory allocations and make judgments about their impact on functional safety.” (October 2023)

Article posted in Research