5 min read

This year’s ICRA 2018 conference features interactive sessions, keynotes, exhibitions, workshops, and much more. Following are some of the interesting keynotes on machine learning, robotics, and more.

Note: International Conference on Robotics and Automation (ICRA) is an international forum for robotics researchers to represent their work. It is a flagship conference of IEEE Robotics and Automation Society.

This conference held at the Brisbane Convention and Exhibition Center from the 21st to 25th May, 2018  brings together experts in the field of robotics and automation. The conference includes delegates in the frontier of science and technology in robotics and automation.

Implementing machine learning for safe, high-performance control of mobile robots

Traditional algorithms are designed based on their a-priori knowledge leveraged from the system and its environment. This knowledge also includes system dynamics and an environment map. Such an approach can allow system to work successfully in a predictable environment. However, if the system is unaware of the environment details, it may lead to high performance losses.

In order to build systems that can work efficiently in unknown and uncertain instances, the speaker, Prof. Angela Schoellig, introduces systems that are capable of learning amidst an operation and adapt the behaviour accordingly.

Angela presents several approaches for online, data-efficient, and safety-guaranteed learning for robot control. In these approaches, the algorithms can:

  • Leverage insights from control theory
  • Make use of neural networks and Gaussian processes, which are state-of-the-art and probabilistic learning methods.
  • Take into account any prior knowledge about system dynamics.

The speaker has also demonstrated how using such novel robot control and learning algorithms can be safe and effective in real-world scenarios. You can check Angela Schoellig’s video below on how she demonstrated these algorithms on self-flying and -driving vehicles, and mobile manipulators.

Meta-learning and the art of Learning to Learn

Pieter Abbeel, in his talk about meta-learning (learning to learn) explains how reinforcement learning and imitation learning have been successful in various domains such as Atari, Go, and so on. You can also check out 6 Key Challenges in Deep Learning for Robotics by Pieter Abbeel presented at the NIPS 2017 conference.

Humans have a default potential to learn from past experiences and can learn new skills far more quickly than machines. Pieter explains some of his recent experiments on meta-learning, where agents learn imitation or the reinforcement learning algorithms and using the algorithms as base can learn from past instances just like humans. Due to meta learning, machines can now acquire any skill just by having a single demonstration or few trials. He states that meta-learning can be applied to general examples such as omniglot and mini-imagenet, which are standard few-shot classification benchmarks.

To know about meta-learning from the ground up, you can check out our article, What is Meta Learning?. You can also read our coverage on Pieter Abbeel’s accepted paper at the ICLR 2018.

Robo-peers: Robust Interaction in Human-Robot Teams

Richard Vaughan in this keynote explains how robots would behave in natural surroundings, i.e among humans, animals, and other peer robots. His team has worked on behaviour strategies for mobile robots. These strategies enable the robots to have sensing capabilities and also allow them to behave sophisticated like humans and have robust interactions with the world and other agents around them.

Richard further described certain series of vision-mediated Human-Robot Interactions conducted within groups of driving and flying robots. The mechanisms used were simple but highly effective.

Form Building Robots to Bridging the Gap between Robotics and AI

Robots posses smart, reactive and user-centered programming systems using which they can physically interact with the world. In current scenarios, every layman is capable of using cutting-edge robotics technology for complex tasks such as force-sensitive assembly and safe physical human-robot interaction. Franka Emika’s Panda, the first commercial robot system, is an example of of a robot with such abilities.

Sami Haddadin, in this talk offers to bridge the gap between model-based nonlinear control algorithms and data-driven machine learning via a holistic approach. He explains that neither pure control-based nor end-to-end learning algorithms are a close match to human-level general purpose machine intelligence. Two recent results reinforce this statement:

i.) Learning of exact articulated robot dynamics by using the concept of first order principle networks.

ii.) Learning human-like manipulation skills by combining adaptive impedance control and meta learning

Panda was, right from the beginning, released with consistent research interfaces and modules to enable the robotics and AI community to build on the developments in the field until then and to push the boundaries in manipulation, interaction and general AI-enhanced robotics. Sami believes this step will positively enable the community to address the immense challenges in robotics and AI research.

Socially Assistive Robots: The Next-Gen Healthcare Helpers

Goldie Nejat puts down her concern by stating that the world’s elderly population is rising and so is dementia, a disease with hardly any cure. She says that robots here, can become a unique strategic technology. She further adds that they can become a crucial part of the society by helping the aged population in their day-to-day activities.

In this talk she presents intelligent assistive robots, which can be used to improve the life of the older populations. The population also includes those suffering from dementia. She discusses how the assistive robots, Brian, Casper, and Tangy socially have been designed to autonomously provide cognitive and social interventions. These robots also help with activities of daily living, and lead group recreational activities in human-centered environments. These robots can serve as assistants to individuals as well as groups of users. They can personalize their interactions as per the needs of the users. These robots can also be integrated into everyday lives of other people outside the aged bracket.

Read more about the other keynotes and highlights on robotics on the ICRA’s official website

Read Next

A Data science fanatic. Loves to be updated with the tech happenings around the globe. Loves singing and composing songs. Believes in putting the art in smart.

LEAVE A REPLY

Please enter your comment!
Please enter your name here