AI Courses for Advanced Learners & Innovators

At UDHY – Unlock Digital Horizons for You, we are all about empowering learners to gain real skills, build meaningful projects, and achieve their goals.

Home › AI Courses › AI for Advanced Learners

Advanced AI — Robotics Series · 3 Courses

You have learned what AI is. You have run your first machine learning model. Now it is time to build the systems that power real autonomous robots — from scratch, with working code, at production depth.

I have been in this industry for more than a decade — co-founding Moovita, Singapore’s first autonomous vehicle company, and spending years as a Principal Research Scientist at A*STAR developing real-world perception and control systems. The advanced curriculum you are about to enter is built directly from that experience — not textbook theory, but the techniques and architectures that are running in production autonomous systems right now in 2026.

This series covers three courses in a deliberate sequence. Deep learning gives the robot its eyes — the ability to perceive and understand the physical world. Reinforcement learning gives it the ability to act — to learn complex behaviour through trial, reward, and iteration. Autonomous navigation and SLAM gives it the ability to go somewhere — to map unknown environments and plan paths through them in real time. Together, these three courses form the complete perception → action → navigation pipeline that underpins every serious autonomous system operating today.

Every course includes working Python code you can run immediately, real-world robotics examples drawn from Waymo, Boston Dynamics, and Amazon Robotics, and internal connections to UDHY’s growing library of autonomous vehicle analysis, sensor fusion guides, and robotics course series. All three are completely free.

Before you start — are you in the right place?

This series assumes you are comfortable with Python, understand what a neural network is, and have completed (or can test out of) UDHY’s Machine Learning Fundamentals. If you are newer to AI, start with Introduction to AI — it is free and takes under 2 hours.

What You Will Build Across This Series

COURSE 1

A working CNN-based object classifier in PyTorch — the perception backbone of every serious robot system

COURSE 2

A PPO agent that learns bipedal locomotion from scratch — the same algorithm behind Boston Dynamics’ Atlas

COURSE 3

A working SLAM system and A* path planner — the navigation core of every commercial autonomous robot

The Three Advanced Courses — At a Glance

Course 1 of 3

Deep Learning for Robotics: From Neural Networks to Autonomous Systems

Before a robot can act, it must perceive. This course teaches you the deep learning architectures that power robot vision — CNNs that classify objects at 200+ FPS, LSTMs that predict pedestrian trajectories, and the sim-to-real transfer techniques that make simulation training economically viable. You will implement a complete PyTorch perception pipeline and understand exactly why Waymo, Boston Dynamics, and Amazon Robotics built their systems the way they did.

CNNs & PyTorch, Transfer Learning, Sim-to-Real Transfer, Domain Randomisation & LSTM & Transformers

⏱ 10–14 hours · Self-paced📋 5 modules💻 2 code projects

What you will build: A 3-layer CNN trained on CIFAR-10 in PyTorch, deployable on Jetson AGX Orin at 200+ FPS. Understand the YOLOv8/v11 family, ResNet-50 transfer learning, and NVIDIA Isaac Sim for domain randomisation.

Start Course 1

Course 2 of 3

Reinforcement Learning for Robotics: Teaching Machines to Act

Perception is only half the problem. This course teaches you how robots learn to act — through reward, iteration, and the mathematical framework of Markov Decision Processes. You will implement Q-Learning from scratch, understand Deep Q-Networks, and train a PPO agent that learns bipedal locomotion from zero — the same algorithm OpenAI used to train the Rubik’s Cube hand and ETH Zurich used for ANYmal. The course closes with imitation learning: how robots learn from human demonstrations rather than random exploration.

Markov Decision Processes, Q-Learning, Deep Q-Networks, PPO & Imitation Learning

⏱ 12–16 hours · Self-paced📋 5 modules💻 2 code projects

What you will build: A Q-Learning agent achieving 95%+ on FrozenLake navigation, then a PPO-trained BipedalWalker agent using Stable-Baselines3. Understand how Boston Dynamics, DeepMind’s Gemini Robotics, and commercial delivery robots were all trained.Start Course 2 →

Start Course 2

Course 3 of 3

Autonomous Navigation and SLAM: How Robots Find Their Way

A robot that can see and act still needs to know where it is and how to get somewhere. SLAM — Simultaneous Localisation and Mapping — solves the fundamental circular problem of robot navigation: to build a map you need to know where you are, and to know where you are you need a map. This course teaches you the probabilistic mathematics of SLAM, implements ICP scan matching and A* path planning in Python, and shows you how to configure ROS 2 Nav2 — the production navigation stack used by Amazon Robotics, Clearpath, and Fetch Robotics today.

SLAM Mathematics, ICP Scan Matching, A* Path Planning, ROS 2 Nav2 & Semantic SLAM

⏱ 12–16 hours · Self-paced📋 5 modules💻 3 code projects

What you will build: A complete 2D LiDAR SLAM system with ICP scan matching and occupancy map generation, an A* path planner navigating a grid with obstacles, and a Nav2 waypoint navigation client for multi-stop autonomous routing in ROS 2.Start Course 3 →

Start Course 3

Where This Fits in the UDHY Learning Path

Start here

AI for Beginners
Intro · Tools · ML

Next step

AI for Experts
VLA · MAS · Security

Parallel track

Robotics Courses
ROS 2 · HW· VLA

Read These Before or Alongside This Series

The advanced courses are richer if you have read UDHY’s analysis articles on the real-world systems you are building towards. These are free — and written at practitioner depth:

Why Self-Driving Cars Still Fail — the edge case problem your deep learning model must solve (directly relevant to Course 1)
Sensor Fusion Explained: Cameras, LiDAR & Radar — the multi-sensor perception layer your CNN operates within (Course 1)
Is AI Speeding Up or Slowing Down Autonomous Vehicle Development? — the industry context your RL and SLAM work fits into (Courses 2 & 3)
Level 3 vs Level 4 Autonomy — the deployment target that defines what “good enough” means for the systems you are building (Course 3)
The Complete Guide to AV Teleoperation — the human-in-the-loop architecture that bridges the RL and navigation courses
The Data Gap Threatening the Humanoid Robot Revolution — why the training data you collect in Course 2 matters so much

Dr. Dilip Kumar Limbu — Series Author

Designed by Dr. Dilip Kumar Limbu — Former Principal Research Scientist, A*STAR · Co-Founder, Moovita, Singapore’s first autonomous vehicle company · 30 years building real-world autonomous systems. UDHY.com.