Deep Perception for Manipulation¶
Deep Pose Estimation¶
Learning-Based Methods¶
- CNN architectures for 6-DOF pose estimation
- Training on synthetic and real data
- Handling symmetry and occlusion
Loss Functions¶
- Direct regression (L2 loss)
- Keypoint-based
- Pose graph optimization
Grasp Detection¶
GGCNN¶
Real-time grasp detection network.
6-DOF Grasp Detection¶
Detect grasp pose including rotation and depth.
Semantic Understanding¶
- Instance segmentation
- Scene graphs
- Task-oriented perception
