Dingding Zheng

Senior Autonomous Driving Software Engineer, Mercedes-Benz

I am a senior autonomous driving software engineer at Mercedes-Benz. Prior to that, I was a graduate student at the GRASP Laboratory, University of Pennsylvania.

At Penn, I worked on DCIST, a project focused on multi-agent and heterogeneous systems, supervised by Prof. Vijay Kumar and Dr. Siddharth Mayya.

My interests lie in embodied intelligence, micro unmanned aerial vehicles, robot perception, and learning. Much of my work focuses on making algorithms safe, agile, and reliable on real robotic platforms.

Experience

Senior Autonomous Driving Software Engineer

Mercedes-Benz · Shanghai, China

Jan 2023 — Present

Principal of Shanghai

UPenn VC Club Summer Trek · Shanghai, China

Jul 2023 — Aug 2023

Autonomous Driving Software Engineer

SenseTime · Shanghai, China

Sep 2021 — Dec 2022

Teaching Assistant

University of Pennsylvania · Philadelphia, PA

Feb 2020 — May 2021

MEAM 620: Advanced Robotics (Fall 2021)

Service

University of Pennsylvania

2022

UPenn Alumni Interviewer

Projects

Unitree G1 humanoid robot performing learned locomotion

G1 Motion Studio

GPU-accelerated motion control for the Unitree G1, combining score-matching motion priors, reinforcement learning, and terrain-adaptive control for natural full-body locomotion.

code

Unitree A1 robot navigating a parkour course

Unitree A1 Parkour

Vision-assisted parkour locomotion for the Unitree A1, trained with PPO in NVIDIA Isaac Gym using procedurally generated obstacles and teacher-to-student depth-policy distillation.

video code

End-to-End Autonomous Drone Racing

A PPO policy trained in NVIDIA Isaac Lab flies a Crazyflie through a six-gate circuit using gate-aware rewards, multi-frame observations, and domain randomization for sim-to-real robustness.

video code

Garment Folding with Vision-Language-Action Models

A garment-folding system built on the Pi0.5 vision-language-action model, combining imitation learning with behavioral cloning and DAgger for robust manipulation.

video code

Robotic Piano Playing

A complete perception, planning, and control framework that enables a robot to detect keys and play musical sequences on a piano.

video code

Quadrotor Control

Trajectory generation and control algorithms for a quadrotor flying aggressive maneuvers.

project page video code

F1TENTH Autonomous Racing

Maximum-speed path and velocity-profile generation using CMA-ES, pure-pursuit control, and obstacle avoidance with ODG-PFM and RRT*.

project page code

SMORES-EP Modular Robots

Controller improvements, test-environment development, and topology-similarity analysis for a reconfigurable modular robot platform.

project page

Human–Robot Interaction in Multi-Agent Traffic

Car dynamics and MPC-based human decision models integrated into a multi-agent environment, with robot agents trained using MADDPG and human agents modeled through social forces.

code

Humanoid Robot SLAM

2D occupancy-grid mapping and localization using laser scans, IMU, and odometry, extended with Kinect imagery to build a textured map.

Visual-Inertial Panorama

A Kalman filter estimates 3D orientation from IMU data, calibrated against Vicon ground truth, to generate real-time panoramas.

Mask R-CNN

Object instance segmentation experiments using the Mask R-CNN framework.

Clipped PPO for Car Racing

Clipped Proximal Policy Optimization applied to OpenAI Gym's CarRacing environment.

Image Super-Resolution with GANs

A generative adversarial network that reconstructs high-resolution images from downsampled inputs.

YOLO Object Detection

A YOLO v1 implementation for detecting objects in street-scene images.

GMM-Based Barrel Detection

A Gaussian mixture model detects barrels in images and estimates their positions in world coordinates.

Object Pose Estimation

A heatmap-based neural network estimates object keypoints and recovers an oilcan's pose from images.

Planar Tracking and Augmented Reality

Homography estimation maps video frames onto logo points for stable planar augmentation.

Difference of Gaussians Feature Detection

A scale-space feature detector approximating the Laplacian of Gaussian with Difference of Gaussians.

Actor–Critic for Acrobot

An actor–critic neural network trained to control the OpenAI Gym Acrobot environment.