MODULE//AI_MODEL_TRACKER
Robotics AI foundation modelsMODELS: 12
MODEL_REGISTRY|12 models
SORT:
| ID | MODEL | ORG | TYPE | PARAMS | TASKS | BENCH | OPEN | STATUS |
|---|---|---|---|---|---|---|---|---|
| AIM-006 | PaLM-E | VLM | 562B | reasoning | 95 | — | Research | |
| AIM-010 | SAM 2 | Meta | Vision | 312M | segmentation | 94 | ✓ | Active |
| AIM-002 | RT-2 | Google DeepMind | VLA | 55B | manipulation | 92 | — | Research |
| AIM-011 | DINO v2 | Meta | Vision | 1.1B | features | 91 | ✓ | Active |
| AIM-003 | π0 (pi-zero) | Physical Intelligence | VLA | 3B | manipulation | 88 | — | Research |
| AIM-009 | CLIP | OpenAI | Vision | 428M | perception | 88 | ✓ | Active |
| AIM-007 | GPT-4V + CaP | OpenAI / Stanford | VLM+Code | — | planning | 85 | — | Research |
| AIM-001 | OpenVLAGOLDEN | Stanford / Berkeley / CMU | VLA | 7.5B | manipulation | 82 | ✓ | Active |
| AIM-005 | RoboCat | Google DeepMind | VLA | 1.2B | manipulation | 78 | — | Research |
| AIM-012 | PointNet++ | Stanford | 3D | 1.4M | point cloud | 78 | ✓ | Classic |
| AIM-004 | Octo | UC Berkeley | VLA | 93M | manipulation | 76 | ✓ | Active |
| AIM-008 | SayCan | Google / Everyday Robots | LLM+Skills | 540B | planning | 74 | — | Deprecated |
RESEARCH_LABS
Google DeepMind Robotics
Foundation models, manipulation
48
papers
5
models
Physical Intelligence
Embodied AI, VLAs
3
papers
2
models
Stanford IRIS Lab
Manipulation, learning
62
papers
4
models
UC Berkeley BAIR
RL, imitation learning
85
papers
6
models
MIT CSAIL
Dexterous manipulation
72
papers
3
models
CMU Robotics Institute
Locomotion, planning
94
papers
5
models
Meta FAIR
Vision, embodied AI
38
papers
4
models
NVIDIA Research
Sim-to-real, Isaac
45
papers
3
models
KEY_DEVELOPMENTS
2024VLA models achieve human-level manipulation on standard benchmarks
2024Open-source VLA models enable transfer to real robots
2023RT-2 demonstrates emergent reasoning in physical tasks
2023LLM-based planning enters robotics mainstream
OPEN_SOURCE_INDEX
Open Source Models6
Closed/Research Only6
50% OPEN SOURCE