Chenggang Liu

About

A curious learner and a passionate explorer in the fields of AI, robotics, and automation

2025 – Present

Motional AD

Working on ML-based trajectory planning for L4 self-driving cars.

2023 – 2025

Meta Reality Lab Research

Worked on the Codec Avatar project, generating photo-realistic avatars. Led the calibration production and developed a fully automated capture system calibration pipeline and software.

2021 – 2023

Aurora Innovation

Worked on decision making and trajectory sampling for L4 self-driving cars. The challenge of sample efficiency led me to ML-based approaches.

2015 – 2021

Uber ATG

Worked on trajectory generation, solving three major challenges: 1) runtime performance 2) behavior formulation and tuning and 3) cost function learning. Created the first Inverse RL training pipeline and demonstrated its effectiveness.

2013 – 2015

Robotics Institute — Carnegie Mellon University

Worked on full-body motion planning and control for humanoid robots. Participated in the DARPA Robotics Challenge with Team WPI-CMU, leading the egress task. The team completed the task in under four minutes with a 100% success rate and were the only Atlas team that never fell.

2008 – 2010

Robotics Institute — Carnegie Mellon University

Pioneered the use of optimal trajectory library for robot planning and control. The trajectory optimization method I used, called DDP and its variants, later became the most popular trajectory optimization method.

Earlier

S3 Graphics

Worked on GPU (Graphics Processing Unit) logic design, which sparked my interest in harnessing computational power to drive advanced AI.