Robotic Manipulation: Perception, Planning, and Control
- Erogato da: MIT
- Docente: Russ Tedrake
- Livello: advanced
- Formato: video-lectures
- Durata: One Semester
- Categoria: manipulation
- Lingua: English
- Prezzo: Gratuito
- Tag: manipulation, perception, motion-planning, deep-learning, kinematics, control
Questo corso introduce gli approcci algoritmici fondamentali per creare sistemi robotici in grado di manipolare autonomamente oggetti fisici in ambienti non strutturati come abitazioni e ristoranti. Gli argomenti comprendono percezione (approcci basati su deep learning e geometria 3D), pianificazione (cinematica dei robot, generazione di traiettorie, pianificazione del movimento senza collisioni, pianificazione di compiti e movimenti, pianificazione in condizioni di incertezza), dinamica e controllo (basati su modello e sull'apprendimento). Gli studenti costruiscono uno stack software che consente a un braccio robotico di manipolare autonomamente oggetti in scene ingombre.
Perché seguirlo
Taught by Russ Tedrake at MIT, this course represents the cutting edge of robotic manipulation research. Unlike introductory courses, it tackles the hard unsolved problems: how do you get a robot to autonomously handle objects in cluttered, unpredictable environments like homes and kitchens? The course combines perception, planning, and control into a complete manipulation stack that students actually build and run.
Cosa imparerai
- Deep learning approaches for robotic perception and object recognition
- 3D geometry for manipulation: point clouds, pose estimation, grasp planning
- Motion planning in cluttered environments with collision avoidance
- Task-and-motion planning for complex multi-step manipulation tasks
- Model-based and learning-based control strategies
- Building a complete autonomous manipulation software stack
Programma
A semester-long course combining lectures with a cumulative software project. Students progressively build a manipulation stack: starting with perception (processing 3D sensor data), then planning (generating collision-free trajectories), and finally control (executing plans on a simulated robotic arm). The course uses Drake simulation environment throughout.
Argomenti principali
- Geometric perception: point clouds, pose estimation
- Deep learning for perception: object detection, segmentation
- Kinematics and trajectory optimization
- Collision-free motion planning
- Grasp planning and force closure analysis
- Task-and-motion planning (TAMP)
- Planning under uncertainty
- Learning-based manipulation policies
Prerequisiti
- Strong programming skills (Python)
- Linear algebra and basic optimization
- Familiarity with deep learning fundamentals
- Prior exposure to robotics concepts recommended
A chi si rivolge
- Graduate students in robotics or computer science
- Researchers working on manipulation and grasping
- Engineers building real-world robotic manipulation systems
- Advanced practitioners who want to understand the full manipulation pipeline