DreamX

DreamX: AMAP Spatial Intelligence Models and Systems

I lead DreamX at Alibaba Group, a 100+ member product-facing AI team building spatial intelligence models and systems for AMAP’s mapping, mobility, local-service, content, interactive-world, and embodied-AI scenarios.

One mission: Spatial Intelligence. We build systems that understand and predict, generate and simulate, plan and act in the real world. DreamX connects academic rigor, open-source reproducibility, engineering discipline, and real-world product deployment.

100+ product-facing AI team members

40+ open-source DreamX projects

50+ recent top-venue papers from DreamX

300M+ users served by AMAP products every day

DreamX System Map

DreamX turns the spatial-intelligence mission into six connected model and system families — DreamX-Predictor, DreamX-World, DreamX-Agent, DreamX-Phi, DreamX-REC, and DreamX-Creator — supported by one shared foundation.

Predictor · REC

Understand & Predict

Connect maps, vision, language, mobility, urban scenes, user intent, and product signals to understand spatial context and forecast how the world evolves.

MobilityBench · Thinking-with-Map · DSFNet · IntTravel

World · Creator

Generate & Simulate

Create and simulate map-native assets, media, 3D scenes, digital content, and persistent interactive worlds with control and spatial consistency.

DreamX-World · Code2World · FluxText · RL3DEdit

Agent · Phi

Plan & Act

Build agents and decision systems that reason, use tools, plan, self-reflect, and turn human goals into digital and physical actions.

LongHorizon-Harness · SkillClaw · AutoDrive-R2 · GPG · CoEvolve

Shared Foundation

Models, Data & Evaluation

Spatial data and knowledge, multimodal models, world modeling, reinforcement learning, generative modeling, infrastructure, and rigorous evaluation.

SpatialGenEval · Omni-WorldBench · FASA · USP · RealQA

Open Source

DreamX open-source releases are hosted in the AMAP-ML GitHub organization. The organization name preserves the established URL, while DreamX is the public team and portfolio brand. Together with earlier releases including YOLOv6 and MobileVLM, the work has accumulated 10,000+ GitHub stars.

Project	Stars	Description
SkillClaw		Collective skill evolution for AI agents
FluxText		Controllable scene-text editing and visual asset generation
Code2World		GUI world model via renderable code generation
DreamX-World		Interactive world model for virtual environments
GPG		Simple and strong RL baseline for reasoning
MobilityBench		Route-planning agent benchmark for real-world mobility
YOLOv6		Industrial-grade real-time object detection
MobileVLM		Real-time mobile vision-language model

Join DreamX

We are looking for talented interns, full-time researchers, and AI engineers who want to build research-grade AI systems grounded in real map, mobility, local-service, and interactive AI scenarios.

Spatial intelligence LLM agents Reinforcement learning World models Multimodal learning Embodied AI Recommendation Generative AI

What We Offer

Competitive compensation and benefits
Publish at top venues while building systems with real product paths
Work on large-scale training, evaluation, open-source releases, and deployment
Access to AMAP scenarios in maps, mobility, local services, and interactive AI
Collaborate across research, engineering, and product teams
Grow in a team where strong first-author work, system ownership, and real product impact are all valued

What We Look For

Strong background in ML, CV, NLP, RL, agents, multimodal learning, or generative AI
Publications at top venues (preferred but not required for interns)
Solid coding skills and engineering mindset
Strong research taste and the ability to turn ideas into reproducible systems
Interest in AI systems that are evaluated by both academic rigor and real users

Interested? Send your CV, representative publications or projects, and the direction you want to work on to cxxgtxy@gmail.com.

Recommended subject: DreamX Application - Your Name - Direction.

Xiangxiang Chu（初祥祥）