Humanoids/Legged Robots

• Datasets:
  • InternData-H1: The largest open-sourced 3D human motion dataset with text annotation, including 2.5k hours 1.9M episodes.
• Models and Research:
  • UniHSI: Unified Human-Scene Interaction via Prompted Chain-of-Contacts
  • HIMLoco: Hybrid Internal Model: Learning Agile Legged Locomotion with Simulated Robot Response
  • 🏆HoST [Best Systems Paper Finalist at RSS 2025]: Learning Humanoid Standing-up Control across Diverse Postures
  • HOMIE: Humanoid Loco-Manipulation with Isomorphic Exoskeleton Cockpit

Manipulation

• Datasets:
  • InternData-A1: A hybrid synthetic-real manipulation dataset integrating 5 heterogeneous robots, 15 skills, and 200+ scenes, emphasizing multi-robot collaboration under dynamic scenarios.
  • InternData-M1: A large-scale synthetic dataset for generalizable pick-and-place over 80K objects, with open-ended instructions covering object recognition, spatial and commonsense reasoning, and long-horizon tasks.
• Models and Research:
  • InternVLA-A1: Unifying Understanding, Generation, and Action for Robotic Manipulation
  • InternVLA-M1: A Spatially Guided Vision-Language-Action Framework for Generalist Robot Policy
  • F1-VLA: Visual foresight generation for planning-based control
  • VLAC: A generalist vision-language-action-critic model for robotic real-world reinforcement learning
  • Seer: Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation
  • RoboSplat: Novel Demonstration Generation with Gaussian Splatting Enables Robust One-Shot Manipulation
  • GenManip: LLM-driven Simulation for Generalizable Instruction-Following Manipulation

Navigation

• Datasets:
  • InternData-N1: A high-quality navigation dataset with the most diverse scenes and extensive randomization across embodiments/viewpoints, including 3k+ scenes and 830k VLN data.
• Models and Research:
  • InternVLA-N1: An Open Dual-System Vision-Language Navigation Foundation Model with Learned Latent Plans
  • NavDP: Learning Sim-to-Real Navigation Diffusion Policy with Privileged Information Guidance
  • StreamVLN: Streaming Vision-and-Language Navigation via SlowFast Context Modeling
  • VLN-PE: A Holistic Study of Physical and Visual Disparities in Vision-and-Language Navigation

AIGC for Embodied AI

• Datasets:
  • OmniWorld: A large-scale, multi-domain, multi-modal dataset, enables significant performance improvements in 4D reconstruction and video generation.
• Models and Research:
  • MeshCoder: Generate Structured 3D Object Blender Code from Point Clouds
  • Infinite-Mobility: Scalable High-Fidelity Synthesis of Articulated Objects via Procedural Generation
  • Aether: Geometric-Aware Unified World Modeling

3D Vision and Embodied Perception

• EmbodiedScan: A Holistic Multi-Modal 3D Perception Suite Towards Embodied AI
• 🏆PointLLM [Best Paper Candidate at ECCV 2024]: Empowering Large Language Models to Understand Point Clouds
• MMSI-Bench: A Benchmark for Multi-Image Spatial Intelligence
• OST-Bench: Evaluating the Capabilities of MLLMs in Online Spatio-temporal Scene Understanding

3D Assets for Embodied AI

• InternScenes: A large-scale interactive indoor scene dataset with realistic layouts, 40,000 diverse scenes and 1.96M 3D objects.