This repository supports a course designed to explore the theoretical and practical advances necessary to build intelligent robots capable of safely operating in dynamic environments such as homes, hospitals, and factories. The course focuses on the intersection of Robotics, Control, and Machine Learning, addressing current challenges and state-of-the-art approaches to enable robots to handle unstructured, real-world problems.
Through this repository, you will:
- Understand the current challenges in robotics and control.
- Gain familiarity with the main AI/ML paradigms applied in robotics.
- Develop the ability to critically read state-of-the-art robotics literature.
- Learn to select appropriate AI/ML approaches for tackling unstructured problems in robotics.
This repository is organized into folders, each corresponding to a key topic covered in the course. Below is a brief description of each folder with links to their contents:
- Introduction to Machine Learning: Covers the fundamentals of machine learning, including supervised, unsupervised, and reinforcement learning paradigms used in robotics.
- Link to Folder/script: Introduction to Machine Learning.
- Introduction to Deep Learning: Explores deep learning techniques, such as neural networks, and their applications in robotic perception and decision-making.
- Link to Folder/script: Introduction to Deep Learning.
- Introduction to Control: Introduces control theory, including PID and LQR control, applied to systems like the cartpole inverted pendulum using MuJoCo.
- Link to Folder/script: Introduction to Control.
- Introduction to Reinforcement Learning: Provides an overview of reinforcement learning concepts, focusing on how agents learn optimal behaviors through interaction.
- Link to Folder/script: ntroduction to Reinforcement Learning.
- DQN: Demonstrates Deep Q-Networks (DQN), a reinforcement learning algorithm combining Q-learning with deep neural networks for robotic tasks.
- Link to Folder/script: Deep_Q_learning_Network.
- Model Predictive Control: Explores Model Predictive Control (MPC), an advanced control technique for optimizing robot trajectories in real-time.
- Link to Folder/script: Model Predictive Control.
- Model-Based Reinforcement Learning: Introduces model-based reinforcement learning approaches, leveraging system models to improve learning efficiency in robotics.
- Link to Folder/script: Model-Based Reinforcement Learning.
- Perception: Covers perception techniques, including computer vision and sensor processing, critical for robots to interpret their environments.
- Link to Folder/script: Perception.
- Policy Gradients: Policy gradients are a fundamental class of reinforcement learning algorithms that directly optimize an agent's policy parameters by estimating the gradient of expected cumulative rewards, enabling effective learning in complex, high-dimensional action spaces.
- Link to Folder/script: Policy Gradient.
- Bayesian Optimization: Bayesian optimization is an efficient method for optimizing expensive black-box functions by constructing a probabilistic surrogate model, often a Gaussian process, to approximate the objective. It iteratively selects promising points using acquisition functions that balance exploration of uncertain areas and exploitation of known high-performing regions.
- Link to Folder/script: Bayesian Optimization.
- RL from Images: Reinforcement learning from images is a subfield of machine learning that focuses on training agents to make decisions based on visual input. In this approach, the agent directly learns a policy or value function from raw image data, without the need for handcrafted features or explicit state representations. This allows the agent to perceive and understand the environment solely through visual observations.
- Link to Folder/script: RL from Images.
- Imitation Learning: Imitation learning is a machine learning approach where an agent learns to perform a task by imitating the behavior of an expert. In imitation learning, the agent learns from a dataset of expert demonstrations, where each demonstration consists of a sequence of states and the corresponding actions taken by the expert. The goal of imitation learning is to learn a policy that can mimic the expert's behavior and perform the task accurately.
- Link to Folder/script: Imitation Learning.
For questions or suggestions, please open an issue on the GitHub repository or contact [shamreen.tabassum@mailbox.tu-dresden.de].