From Machine Learning to Robotics: Challenges and Opportunities for Embodied Intelligence

Abstract

Machine learning has long since become a keystone technology, accelerating science and applications in a broad range of domains. Consequently, the notion of applying learning methods to a particular problem set has become an established and valuable modus operandi to advance a particular field. In this article we argue that such an approach does not straightforwardly extended to robotics -or to embodied intelligence more generally: systems which engage in a purposeful exchange of energy and information with a physical environment. In particular, the purview of embodied intelligent agents extends significantly beyond the typical considerations of main-stream machine learning approaches, which typically (i) do not consider operation under conditions significantly different from those encountered during training; (ii) do not consider the often substantial, long-lasting and potentially safety-critical nature of interactions during learning and deployment; (iii) do not require ready adaptation to novel tasks while at the same time (iv) effectively and efficiently curating and extending their models of the world through targeted and deliberate actions. In reality, therefore, these limitations result in learning-based systems which suffer from many of the same operational shortcomings as more traditional, engineering-based approaches when deployed on a robot outside a well defined, and often narrow operating envelope. Contrary to viewing embodied intelligence as another application domain for machine learning, here we argue that it is in fact a key driver for the advancement of machine learning technology. In this article our goal is to highlight challenges and opportunities that are specific to embodied intelligence and to propose research directions which may significantly advance the state-of-the-art in robot learning.

Developing Intelligent Robots that Grasp Affordance

“…By contrast, interactive perception in animals is necessary for animals to discover the existence and behavior of objects in a world about which they initially know nothing, as in kittens learning to navigate based on their visual experience ( Held and Hein, 1963 ). The real physical world and an agent’s interactions with it have many properties that are incompatible with currently available machine-learning algorithms ( Roy et al, 2021 ).…”

“…One candidate for the comparative gate component would be the thalamocortical loop ( Halassa and Sherman, 2019 ). The system architecture also requires motivational and integrative components not illustrated so that the tolerable level of uncertainty about an entity’s identity can be adjusted to avoid acting precipitously in unfamiliar and potentially dangerous situations ( Roy et al, 2021 ). These might reasonably be performed by the biological basal ganglia projections to thalamus, for which computational neural network models are starting to be developed ( Hazy et al, 2007 ).…”

“…Such hierarchies are in widespread use in perceptual neural networks ( Hinton, 2012 ) and they have been extended to separately trainable modules for perception, policy and action ( Hamalainen et al, 2019 ). Integrating them into physically embodied systems to control exploratory actions via middleware with sensory feedback will be challenging ( Roy et al, 2021 ) but may be a necessary step toward truly intelligent machines.…”

Tree Detection and Diameter Estimation Based on Deep Learning

“…Hence, automation could increase operator efficiency while reducing mental fatigue [Ortiz et al, 2014, Visser andObi, 2021]. A technology driver behind advancements in automation is machine learning, which is an established approach to advance particular fields [Roy et al, 2021]. Therefore, applying learning methods to a problem set specifically to the field of forestry, such as harvesting (Figure 1), forwarding, or navigation, will play an important role in the future of forestry automation.…”

“…However, forests are harsh, unstructured outdoor environments [Thorpe and Durrant-Whyte, 2001]. As an environment becomes less structured, its complexity increases beyond the typical welldefined boundaries and narrow operating range established in structured environments, possibly leading to operational shortcomings of autonomous systems [Roy et al, 2021].…”