Course Outline
Introduction to Robotic Manipulation and Deep Learning
- Overview of manipulation tasks and system components
- Traditional vs. learning-based approaches
- Deep learning in perception, planning, and control
Perception for Manipulation
- Visual sensing and object detection for grasping
- 3D vision, depth sensing, and point cloud processing
- Training CNNs for object localization and segmentation
Grasp Planning and Detection
- Classical grasp planning algorithms
- Learning grasp poses from data and simulation
- Implementing grasp detection networks (e.g., GGCNN, Dex-Net)
Control and Motion Planning
- Inverse kinematics and trajectory generation
- Learning-based motion planning and imitation learning
- Reinforcement learning for manipulation control policies
Integration with ROS 2 and Simulation Environments
- Setting up ROS 2 nodes for perception and control
- Simulating robotic manipulators in Gazebo and Isaac Sim
- Integrating neural models for real-time control
End-to-End Learning for Manipulation
- Combining perception, policy, and control in unified networks
- Using demonstration data for supervised policy learning
- Domain adaptation between simulation and real hardware
Evaluation and Optimization
- Metrics for grasp success, stability, and precision
- Testing under varying conditions and disturbances
- Model compression and deployment on edge devices
Hands-on Project: Deep Learning-Based Robotic Grasping
- Designing a perception-to-action pipeline
- Training and testing a grasp detection model
- Integrating the model into a simulated robotic arm
Summary and Next Steps
Requirements
- Strong understanding of robotics kinematics and dynamics
- Experience with Python and deep learning frameworks
- Familiarity with ROS or similar robotic middleware
Audience
- Robotics engineers developing intelligent manipulation systems
- Perception and control specialists working on grasping applications
- Researchers and advanced practitioners in robot learning and AI-based control
Testimonials (3)
Supply of the materials (virtual machine) to get straight into the excersises, and the explanation of the Ros2 core. Why things work a certain way.
Arjan Bakema
Course - Autonomous Navigation & SLAM with ROS 2
Well-explained examples of exercises by the trainer
Mariusz - Politechnika Opolska
Course - Artificial Intelligence (AI) for Mechatronics
Machine Translated
its knowledge and utilization of AI for Robotics in the Future.
