Autonomous driving software development heavily relies on algorithm parameter tuning at scale. Hand-tuning in simulation is a common practice, but it can be time-consuming, error-prone, and not scalable to various complex driving scenarios or large parameter search spaces. At Zoox, we have developed an autotuning platform that accelerates algorithm development by leveraging large-scale, distributed simulation and metrics evaluation. This talk will cover how we utilized Ray to scale simulation and metrics workloads at Zoox, and demonstrate our autotuning process that allows developers to improve autonomous driving without changing any code. Attendees will gain insights into how Ray's capabilities, such as scalability and fault tolerance, are used in our platform. We will also highlight some of the key lessons we learned while developing and deploying our autotuning platform, and provide a glimpse into the future of metrics-driven algorithm development.
Yunpeng Pan is a Staff Software Engineer at Zoox, where he leads a software development team building Zoox's Autotuning Platform. His team focuses on distributed testing and validation infrastructure, large-scale derivative-free optimization, and their integration and deployment into self-driving software development processes. Prior to Zoox, he held engineering positions at Lyft, Magna, and JD.com. His interests are motion planning, machine learning, optimization, and their applications to autonomous driving.
Ritwik Bera is a Software Engineer in the Planning and Control team at Zoox, Inc. At Zoox, Ritwik works on platform stability and core architecture of the company's distributed planner evaluation framework. The framework validates every piece of code that goes into the planning and decision-making stack that is deployed on Zoox's self-driving vehicles. Prior to Zoox, Ritwik earned his MS in Aerospace Engineering from Texas A&M University, College Station.
Come connect with the global community of thinkers and disruptors who are building and deploying the next generation of AI and ML applications.