|
|
--- |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
|
|
|
# MaNI Series |
|
|
``` |
|
|
The MaNI series is a real-world simulation dataset for robotic manipulation, focusing on complex |
|
|
human-object interaction and fine-grained motion modeling. This dataset systematically covers various manipulation |
|
|
types common in real-world robotic operation scenarios, including grasping, pushing, pulling, rotating, assembly, |
|
|
and tool use. It models real-world contact, friction, constraints, and dynamics through high-fidelity physical simulation. |
|
|
MaNI employs a multi-modal data acquisition and synchronization mechanism, including RGB/Depth visual observations, |
|
|
6-DoF poses of objects and end-effectors, joint states, force/torque feedback, and continuous motion trajectories, |
|
|
providing rich and stable data support for learning end-to-end manipulation strategies from perception to control. |
|
|
Unlike purely simulated or purely real-world datasets, MaNI reduces the sim-to-real gap by reconstructing real-world |
|
|
scenarios and calibrating parameters in the simulation environment. This dataset is particularly suitable for research |
|
|
directions such as imitation learning, reinforcement learning, and manipulation planning, and can serve as an important |
|
|
benchmark for robotic skill learning, generalization ability evaluation, and complex interaction reasoning. |
|
|
``` |
|
|
|
|
|
MODAL : Data Confuser "Small files only, Big files confusion" |
|
|
|
|
|
CHARS : 517253f76137x0p |
|
|
|
|
|
DCVERSION : 0.0.1.2 built 260210 |
|
|
|
