Jicong Ao
I am a Research Engineer in the Humanoid Robot Group at the AI Institute of Chinatelecom (TeleAI) in Shanghai. My research focuses on robot learning and task and motion planning, with current work on simulation foundations for robot manipulation.
At TeleAI, I lead the simulation team developing scene, articulated-object, task, and demonstration generation for manipulation data production and sim-to-real policy transfer. My experience also includes agentic vision-language-action model deployment, robot control systems, LLM-based behavior tree generation, robot assembly, and robot milling.
Education
M.Sc. in Mechanical Engineering
Fields: robot learning and task and motion planning
B.Sc. in Industrial Engineering
Field: production planning and optimization
Work Gallery
TeleArt
Large-scale simulation and synthetic pretraining for zero-shot articulated-object manipulation across real robot platforms.
ProjectTeleHive
A scalable distributed data synthesis system for large-scale robot manipulation demonstration synthesis with high parallelism.
ProjectTeleIllusion
A high-parallelism Isaac Lab simulation platform for skill-based motion generation, domain randomization, and large-scale robot manipulation data synthesis.
ProjectMultimodal Assembly Reasoning
Spatial reasoning and pose prediction for multimodal robotic assembly.
ProjectSDGScene
User-intent-driven indoor scene generation via semantic dependency graphs, combining VLM reasoning with constraint-based optimization.
ProjectMulti-Step Task Planning
Ontology-based planning and scheduling for a large-scale assembly production line.
ProjectWittenstein Gearbox Assembly
LLM-generated behavior tree execution for a multi-step gearbox assembly.
ProjectProjects
Large-scale synthetic demonstration generation for zero-shot articulated object manipulation and VLA pretraining.
High-parallelism robot manipulation simulation with domain randomization and cluster scheduling.
A scalable distributed data synthesis system for large-scale robot manipulation demonstration synthesis with high parallelism.
VLM- and constraint-based indoor scene generation.
LLM-based task and demonstration generation for bimanual dexterous manipulation.
LLM- and behavior tree-based robot task planning and execution.
Robot skill composition and learning with runtime tool exchange.
Publications
SDGScenes: User-intent driven indoor scene generation via semantic dependency graph
Pattern Recognition, 2026
Experience
Research Engineer, Humanoid Robot Group
Research Intern, Embodied AI Group
Research Intern & Thesis Student, Collective Learning Group
Thesis Student, Robotic Processing Group
Design and Manufacturing Engineer, Technical Department
