Haoran Song

End-to-End Driving Algorithm Lead

I lead end-to-end driving algorithm development at Horizon Robotics, heading a 70+ member R&D team for Horizon SuperDrive (HSD). My work connects deep learning with autonomous driving, from research ideas to production autonomous driving.

CV Email Google Scholar

Experience

Horizon Robotics

2024 – Present

I led two core HSD capabilities from development to production:

DrivePath (DP) enabled the transition from HD-map-based to mapless driving.
DriveTraj (DT) enabled the transition from a cascaded pipeline to one-stage end-to-end driving.

Huawei ADS

2021 – 2024

As a Principal Engineer, I led prediction-model development for ADS 1.0 and 2.0 and initiated an early end-to-end planning prototype.

More about my work, production results, and honors in my CV.

Background

I received my Ph.D. in Robotics from HKUST in 2021, advised by Michael Yu Wang and Qifeng Chen. I also visited KTH in 2018, working with Danica Kragic and Kaiyu Hang, and received my B.Eng. from Harbin Institute of Technology in 2016.

Selected Research & Publications

Prediction and planning for autonomous driving; motion planning for robotics.

	Learning to Predict Vehicle Trajectories with Model-based Planning Haoran Song, Di Luan, Wenchao Ding, Michael Yu Wang, Qifeng Chen Conference on Robot Learning (CoRL 2021), London, UK Ranked 1st on the Argoverse Motion Forecasting Leaderboard as of March 2021 Combining model-based trajectory generation with learned prediction for feasible, robust vehicle trajectories. project page / paper / bibtex
	PiP: Planning-informed Trajectory Prediction for Autonomous Driving Haoran Song, Wenchao Ding, Yuxuan Chen, Shaojie Shen, Michael Yu Wang, Qifeng Chen European Conference on Computer Vision (ECCV 2020), Glasgow, Scotland, UK Using ego-vehicle plans to predict how surrounding agents will respond. project page / paper / bibtex
	Multi-Object Rearrangement with Monte Carlo Tree Search: A Case Study on Planar Nonprehensile Sorting Haoran Song, Joshua A. Haustein, Weihao Yuan, Kaiyu Hang, Michael Yu Wang, Danica Kragic, Johannes A. Stork IEEE International Conference on Intelligent Robots and Systems (IROS 2020), Las Vegas, USA Learning a policy from MCTS-generated data, then feeding it back into MCTS to improve multi-object sorting. project page / paper / bibtex
	Herding by Caging: A Formation-based Motion Planning Framework for Guiding Mobile Agents Haoran Song, Anastasiia Varava, Oleksandr Kravchenko, Danica Kragic, Michael Yu Wang, Florian T. Pokorny, Kaiyu Hang Autonomous Robots (AURO), 2021 Using geometry and topology to plan robot formations that guide mobile agents. paper / video
	Perching and resting - A paradigm for UAV maneuvering with modularized landing gears Kaiyu Hang, Ximin Lyu, Haoran Song, Johannes A Stork, Aaron M Dollar, Danica Kragic, Fu Zhang Science Robotics, 2019 Cover article Enabling UAVs to perch and rest on structures, saving energy while maintaining stable viewpoints. Featured in MIT Technology Review, Nature Electronics, National Public Radio, etc. paper / video / bibtex
	Fingertip surface optimization for robust grasping on contact primitives Haoran Song, Michael Yu Wang, Kaiyu Hang IEEE Robotics and Automation Letters (RA-L), 2018 & IEEE International Conference on Robotics and Automation (ICRA 2018), Brisbane, Australia Reframing robust grasp planning as geometric matching with learned contact geometries. paper / bibtex

* denotes co-first authorship (equal contribution).

Other Research & Patents

	Reinforcement Learning in Topology-based Representation for Human Body Movement with Whole Arm Manipulation Weihao Yuan, Kaiyu Hang, Haoran Song, Danica Kragic, Michael Y. Wang, Johannes A. Stork IEEE International Conference on Robotics and Automation (ICRA 2019), Montreal, Canada paper / bibtex / code
	Robotic fingertip design and grasping on contact primitives Kaiyu Hang, Haoran Song US Patent App. 16/192,169, 2019 patent info / bibtex

Academic Activities

Reviewer: CVPR, ECCV, CoRL, ICLR, ICRA, IROS, IEEE T-RO, and IEEE RA-L.