PERSONAL WEBSITE中文
Aiming

Aiming毛一鸣 · Yiming Mao

Data Lead & Algorithm Expert · Pokebot

From autonomous driving to embodied intelligence.

I have seven years of experience developing autonomous-driving algorithms, spanning perception and prediction and the full workflow from data and model training to deployment. I am now a Data Lead and Algorithm Expert at PokeBot, contributing to RoboAPI development and leading data work and embodied foundation-model pretraining in collaboration with Huazhe Xu.

During my internship at TuSimple, I collaborated with Lue Fan, Feng Wang, and Naiyan Wang on 3D perception research and co-authored CTRL (ICCV 2023 Oral). CTRL achieved SOTA on both the Waymo Open Dataset 3D Detection and 3D Tracking leaderboards in the 2023 paper’s offline evaluation, providing a high-accuracy approach to industrial 3D object auto-labeling.

At Horizon Robotics, I worked with Xuewu Lin and Zhizhong Su and led stereo vision algorithm development, translating research into engineering implementations that supported production deployment.

At LimX Dynamics, I reproduced the towel-folding capability shown in DYNA-1’s public videos, achieving faster folding than the demonstrations in those videos. This research led to ARM, on which I am first author, with code released through FluxVLA.

My work connects algorithm research, data, model training, and deployment to build reliable intelligent systems. If my work interests you, I welcome conversations about research, technical collaboration, and career opportunities.

Education

Hunan University

Master’s student · Autonomous driving

Research team of Academicians Zhihua Zhong and Rongjun Ding

Tsinghua University

Joint master’s training

Research team of Academician Keqiang Li

Intelligent and Connected Vehicles research group; co-advised by Ming Gao.

Selected publications & projects

Robotics · Reward modeling · 3D perception
RoboAPI · Poke-Aurora

Industry project · October 2026

RoboAPI · Poke-Aurora

A robot API powered by the Poke-Aurora world-action model, translating visual observations, robot state, and language instructions into action chunks for zero-shot manipulation across environments and robot configurations.

My contribution
Contributed to RoboAPI development, with responsibility for data and embodied foundation-model pretraining.
Evidence & result
Launched as a public technical preview on October 30, 2026, enabling developers to test the model on their own robots in real environments.
Official overview & demosDocumentationTry RoboAPI
ARM: Advantage Reward Modeling for Long-Horizon Manipulation

CVPR 2026 Workshop · Poster

ARM: Advantage Reward Modeling for Long-Horizon Manipulation

Yiming Mao, Zixi Yu, Weixin Mao, Yinhao Li, Qirui Hu, Zihan Lan, Minzhao Zhu, Hua Chen

Learning relative advantage from three simple labels—progressing, regressing, and stagnant—to guide policy training for long-horizon manipulation.

Authorship
First author
Evidence & result
Reported 99.4% success on long-horizon towel folding with GR00T N1.5 + ARM (AW-BC), under the paper’s evaluation setting.
ProjectPaperCodeCVF
COSA 0.5 · Humanoid VLA V³-0

Industry project · 2026

COSA 0.5 · Humanoid VLA V³-0

The V³-0 stack in COSA 0.5 combines humanoid whole-body mobile manipulation with human-in-the-loop reinforcement learning on real robots.

My contribution
Provided support for data, models, and selected algorithms.
Evidence & result
The official report demonstrates continuous room tidying and whole-body expert intervention on the Oli humanoid.
Official technical report & demos
FluxVLA Engine: A One-Stop VLA Engineering Platform for Embodied Intelligence

arXiv technical report · 2026

FluxVLA Engine: A One-Stop VLA Engineering Platform for Embodied Intelligence

Yinhao Li, Weixin Mao, Zihan Lan, …, Yiming Mao, …, Hua Chen

A modular engineering platform connecting embodied-policy data, training, simulation evaluation, and real-robot deployment.

Authorship
Co-author
Evidence & result
Open-source workflows for distributed training, accelerated inference, and human-in-the-loop correction; evaluation protocols are detailed in the report.
PaperCodeDocumentation
CTRL: Once Detected, Never Lost

ICCV 2023 · Oral

CTRL: Once Detected, Never Lost

Lue Fan, Yuxue Yang, Yiming Mao, Feng Wang, Yuntao Chen, Naiyan Wang, Zhaoxiang Zhang

Track-centric offline LiDAR 3D detection, combining bidirectional tracking and temporal refinement to improve automatic annotation.

Authorship
Co-author
Evidence & result
Achieved SOTA on both the Waymo Open Dataset 3D Detection and 3D Tracking leaderboards (2023 paper results, offline setting).
PaperCodeInstructionsLeaderboard results

Code & project resources

FluxVLA Engine

Open-source infrastructure for embodied-policy data, training, evaluation, and deployment. Includes VLA/WAM integration and inference workflows.

CTRL · Official implementation

The official implementation of CTRL is maintained in tusen-ai/SST. The repository includes training and evaluation instructions.

ARM · Project resources

Project overview, manipulation demonstrations, method details, and experimental results for ARM.

Professional experience

PokeBot

Current

Data Lead & Algorithm Expert

Contribute to RoboAPI development and lead data work and embodied foundation-model pretraining, collaborating with Huazhe Xu on research and development.

LimX Dynamics

Previously

Embodied Foundation Model Algorithm Researcher

Reproduced the towel-folding capability shown in DYNA-1’s public videos, achieving faster folding than the demonstrations in those videos. This research led to ARM, which I first-authored and open-sourced through FluxVLA.

Worked on reinforcement learning and humanoid loco-manipulation, and provided support for data, models, and selected algorithms in COSA 0.5.

COSA 0.5 · Official technical report

Horizon Robotics

Previously

Autonomous-driving Algorithm Researcher

Worked with Xuewu Lin and Zhizhong Su, leading stereo vision algorithm development and supporting deployment in production programs. My work also covered general perception, occupancy, world models, and end-to-end driving. Mentored by Xuewu Lin.

TuSimple

Internship

Autonomous-driving Algorithm Research

Collaborated with Lue Fan, Feng Wang, and Naiyan Wang on 3D perception research, co-authoring CTRL (ICCV 2023 Oral). CTRL achieved SOTA on both the Waymo Open Dataset 3D Detection and 3D Tracking leaderboards in the 2023 paper’s offline evaluation, providing a high-accuracy approach to industrial 3D object auto-labeling.

During my internship, I received guidance from Yan Yan, Feng Wang, and Naiyan Wang.

Xiaomi EV

Internship

Autonomous-driving Algorithm Research

Collaborated with Hongwei on data feedback loops and cloud-based large-model research and development for autonomous driving.

Let’s connect.

Interested in my work? Let’s talk research, technical collaboration, or career opportunities.

GitHub ↗