ROB108 · AI for Intelligent Robots

Categories: Emerging Technology
Wishlist Share

About Course

Explore how modern AI makes robots more capable: reinforcement learning, imitation learning, sim-to-real transfer, language-driven task planning and vision-language-action models, together with the safety shields, validation and human oversight that keep learned behavior trustworthy.

Advance · 8 hours · 4 lessons, a lab, an assessed assignment and a portfolio piece

Who it is for

Learners with robot programming, perception and navigation experience who want to apply learning-based methods responsibly.

Prerequisites

ROB105 (computer vision) and ROB106 (navigation and ROS 2) or equivalent; comfortable Python with NumPy.

What you will be able to do

  • Formulate a robot task as a reinforcement-learning problem with states, actions and a well-designed reward.
  • Train and evaluate a policy with Gymnasium and Stable-Baselines3 and compare it with a behavior-cloning baseline.
  • Explain sim-to-real transfer techniques such as domain randomization and their limits.
  • Design a safety layer that validates learned or language-generated actions before execution.

Lessons

  • Lesson 1 · Reinforcement learning for robots
  • Lesson 2 · Imitation learning and foundation models
  • Lesson 3 · Safety, evaluation and oversight for learned behavior
  • Lesson 4 · Lab and assignment

Lab: Train, imitate and shield a warehouse robot policy

You build a small grid-world warehouse environment in Gymnasium where a robot must reach a pick location while avoiding shelves and a moving person. You train a PPO policy, train a behavior-cloning policy from scripted demonstrations, compare both and then wrap them with a safety shield that blocks actions entering forbidden cells, measuring how often it intervenes.

Free tools: Google Colab (free) with Gymnasium, Stable-Baselines3 and PyTorch (all free and open source)

Assessed assignment

Submit your Colab notebook with code, plots and a results table comparing PPO, behavior cloning and their shielded versions. Explain your reward design and one reward change that produced unwanted behavior. Discuss what would be needed to transfer the policy to a real robot, including domain randomization and real-world testing. Finish with a one-page safety and oversight plan for a warehouse that wants to use a learning-based or language-instructed robot, covering validation, monitoring, human override and incident reporting.

Criterion Weight What excellent work looks like
Environment and reward design 25% The environment is correct and checked; the reward is justified and reward-hacking risks are discussed.
Training and evaluation 30% Both policies are trained and evaluated on unseen layouts with clear, correctly computed metrics.
Safety shield 25% The shield reliably blocks unsafe actions and its intervention rate and effect on performance are analyzed.
Sim-to-real and oversight plan 20% Transfer challenges and the oversight plan are realistic, specific and proportionate.

Portfolio piece

A reproducible learning-for-robotics notebook comparing reinforcement learning and imitation with a safety shield, plus an oversight plan.

Self-paced, 12-month access. Questions are answered in writing through the student panel. A completion record is issued when you finish the course.

Next step

Continue with ROB109 · Final Project: An Autonomous Robot Task in Simulation · 8 hours · $56.00

Continue to the next course

Prefer the whole program? Program in Robotics & Intelligent Automation · $378.00 (10% less than buying the courses separately)

Show More

What Will You Learn?

  • Formulate a robot task as a reinforcement-learning problem with states, actions and a well-designed reward.
  • Train and evaluate a policy with Gymnasium and Stable-Baselines3 and compare it with a behavior-cloning baseline.
  • Explain sim-to-real transfer techniques such as domain randomization and their limits.
  • Design a safety layer that validates learned or language-generated actions before execution.

Course Content

ROB108 · AI for Intelligent Robots (8 h)
Explore how modern AI makes robots more capable: reinforcement learning, imitation learning, sim-to-real transfer, language-driven task planning and vision-language-action models, together with the safety shields, validation and human oversight that keep learned behavior trustworthy.

  • Lesson 1 · Reinforcement learning for robots
  • Lesson 2 · Imitation learning and foundation models
  • Lesson 3 · Safety, evaluation and oversight for learned behavior
  • Lesson 4 · Lab and assignment
  • ROB108 Assessment · AI for Intelligent Robots
Scroll to Top