Skip to main content

The article is aimed at any military officer or non-commissioned soldier with an interest in future technologies. No formal study in the field is required. The article is useful to appreciate the current pace in artificial intelligence designs and to inform realistic future military capabilities.

Reinforced Learning methods can be combined with data-driven specifications of behaviour to execute similar behaviour in the physical simulation of a robotic system. The following is a summary of several research articles drawn from Berkley Artificial Intelligence Research on current progress in robotic learning.

One method of robotic learning combined state-of-the-art techniques in computer vision and reinforcement learning, to simulate characters to learn a diverse repertoire of skills from video clips drawn from You Tube. Given a single monocular video of an actor performing some skill, such as a cartwheel or a backflip, characters were able to learn policies that reproduce skills in physics simulation, without requiring any manual pose annotations.

Another method employed a reinforcement learning paradigm to train an agent to solve an individual task with a manually designed reward. The reinforcement learning algorithms were used to learn multiple different tasks simultaneously, without additional human supervision. For an agent to acquire skills without human intervention, it must be able to set goals for itself, interact with the environment, and evaluate whether it has achieved its goals to improve its behaviour, all from raw observations such as images.

Imitation learning from a video of a human being can take a huge number of demonstrations hence a robot will struggle if there’s only one demonstration presented. Robots can be equipped with the ability to imitate by observing a human through meta-learning. Meta-learning involves the incorporation of prior experience rather than learning each skill completely from scratch. By incorporating prior experience, the robot should also be able to quickly learn to manipulate new objects while being invariant to shifts in domain, such as a person providing a demonstration, a varying background scene, or different viewpoint.

Computer vision, reinforcement learning through reward-based parameters and combining meta-learning with imitation learning allows robotic systems to observe and imitate behaviour within manually designated goals.

00
Cite Article
Harvard
APA
Footnote
RIS
(Bin Peng et al., 2018)
Bin Peng X. et al. 2018. 'Computer Vision, Reinforcement Learning and Imitation Learning in Robotics'. Available at: https://theforge.defence.gov.au/publications/computer-vision-reinforcement-learning-and-imitation-learning-robotics (Accessed: 25 April 2026).
(Bin Peng et al., 2018)
Bin Peng X. et al. 2018. 'Computer Vision, Reinforcement Learning and Imitation Learning in Robotics'. Available at: https://theforge.defence.gov.au/publications/computer-vision-reinforcement-learning-and-imitation-learning-robotics (Accessed: 25 April 2026).
Xue Bin Peng. et al. "Computer Vision, Reinforcement Learning and Imitation Learning in Robotics", The Forge, Published: October 25, 2018, https://theforge.defence.gov.au/publications/computer-vision-reinforcement-learning-and-imitation-learning-robotics. (accessed April 25, 2026).
Download a RIS file to use in your citation management tools.
Defence Technical Social

Technical Mastery

Advanced Specialisation tech-ecat-level2

Comments