cag
Datasets
All datasets matching “cag”CAGUI
CAGUI: Chinese Android GUI Benchmark
A real-world Chinese Android GUI benchmark designed to evaluate GUI agent models on two complementary capabilities:
Grounding – understanding individual GUI components and linking them to semantics.
Agent – planning and executing multi-step actions to complete user goals on Chinese Android apps.
See AgentCPM-GUI for more details.
🌟 Key Features
Aspect
Grounding
Agent
Objective
GUI widgets grounding / OCR text
Follow… See the full description on the dataset page: https://huggingface.co/datasets/openbmb/CAGUI.strands-kitchen-stereo4d
🍳 Kitchen Stereo4D Dataset
Stereo depth + spatial audio pick-and-place dataset generated by strands-robots on NVIDIA Jetson AGX Thor
Overview
This dataset contains synchronized stereo RGB, metric depth maps, 3D point clouds,
and binaural spatial audio from robot manipulation tasks in a kitchen environment.
Metric
Value
Episodes
10
Total Frames
501
Resolution
640×480
Stereo Baseline
50mm
FPS
30
Robot
Franka Emika Panda (simplified)
GPU
NVIDIA Thor… See the full description on the dataset page: https://huggingface.co/datasets/cagataydev/strands-kitchen-stereo4d.cagedkimodo-g1-longhorizon-v3
Kimodo G1 Long-Horizon v3
10,000 episodes • 21.3M frames • 50Hz • 29-DoF Unitree G1
Text-conditioned whole-body humanoid trajectories generated with nvidia/Kimodo-G1-RP-v1 and tracked by cagataydev/protomotions-gtp-unitree-g1. Each episode chains ~50 primitive motions into a single ~3-minute "long-horizon day" of the humanoid living through 7 activity families (warmup → work → greet → patrol → combat → dance → cooldown → spontaneous).
Built with 🦆 strands-labs/robots — the… See the full description on the dataset page: https://huggingface.co/datasets/cagataydev/kimodo-g1-longhorizon-v3.kimodo-g1-longhorizon-v2
Kimodo G1 Long-Horizon v2
10,000 episodes · 9.74M frames · 65 hours of Unitree G1 motion · text-conditioned, long-horizon
The largest open text-conditioned long-horizon locomotion+manipulation dataset for the Unitree G1 humanoid. Each episode is a chained multi-skill sequence (3–8 atomic motions) generated end-to-end from a natural-language prompt via a Kimodo → ProtoMotions GTP → MuJoCo pipeline, then recorded in LeRobot v3.0 format at 50 Hz.
Companion to… See the full description on the dataset page: https://huggingface.co/datasets/cagataydev/kimodo-g1-longhorizon-v2.CAG