The recorded lectures and webinars of the Robot Learning lecture series run by Professor Jitendra Malik at the University of  California, Berkeley have been released as a YouTube playlist.

To understand the concept behind these lectures, here are a couple of quotes:

"Self-driving cars are still not a product - Yet a 16 year old can learn to drive a car in less than 20 hours"

"It is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility"

That is the magic of the human mind; a robot can't still do things that a child can do. But why is that ?

The lecturer thinks that it mainly comes down to the availability and type of digitized data found on the web used for training machine learning models:

We have tons of textual data - books, Wikipedia, reddit, code on Github, etc etc – close to a trillion words that we can use for GPT like models. For training robots we would need data of my internal state – what I perceive, my brain activations, my muscle commands- very personal stuff – and then transfer it to a robot. Not so easy.So what can we do?

The goal of the lecture then is to take us through examples of how we can use machine learning in robotics and teach how to build machines that can emulate the capabilities of humans and other animals at motor control.

Motor control is defined as connecting perception to action in the physical world. This is looked at from different and finer mobility perspectives:

• legs over wheels
• multi-finger hands over parallel jaw grippers
• rich visual and tactile perception over minimal sensing
• humanoid robots in the home over specialist robots on the factory floor

Conceptually it revolves around:

• biological motor control basics for inspiration
• paradigms for robot motor skill acquisition
• case studies of locomotion, navigation and manipulation

and comprises of the following lectures:

Lecture 1A-Introduction
Lecture 1B-Biomechanics of walking and running
Lecture 2A-Robot mechanisms - kinematics and dynamics
Lecture 2B-The human hand and dexterous object manipulation
Lecture 3A-Robot hands
Lecture 3B-Proprioception and tactile perception
Lecture 4A-Vision for action
Lecture 4B-The developmental perspective on motor control
Lecture 5A-Robot dynamics, control, and motion planning
Lecture 5B-Computational neuroscience perspective on prediction and control
Lecture 6AB-Reinforcement Learning
Lecture 7AB-Behavior cloning
Lecture 8AB-Visual Imitation
Lecture 9AB-Case Studies in Locomotion
Lecture 10AB-Case Studies in Navigation
Lecture 11AB-Case Studies in Dexterous Manipulation
Lecture 12AB-Long horizon planning and the role of language

The lecture series comes down to trying to map the cognitive processing and neuromechanics behind human locomotion. That includes topics like the biomechanics of movement, hand anatomy, roles of vision and eye movements, embodied cognition, robot dexterity, kinematics and dynamics, coordinates systems, understanding body motions etc.

Note that it is a graduate level class heavy in maths and requires prerequisite knowledge at the level of Deep Learning (Bishop, Springer), and Reinforcement Learning (Sutton & Barto, MIT Press).

The lectures are available as a Youtube playlist, span between 45 and 90 minutes each, and the reading material can be found on its Github repo. Links below.

In conclusion, whether you're actually in the business of robotics or studying the matter, or you're just someone who wants to understand the mechanics of how humans learn and move and how this translates to the robot world, then this is the right place to be.

More Information

Robots That Learn - UC Berkeley CS 294-277
Main course site on Github

Related Articles

NVIDIA Releases Free Courses On AI    

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info