Robots and A.I.

“Human Robot Interaction” is the first step to understanding how robots assist humans with tasks. Building robots that provide value to humans means first understanding the signals human beings project into the world and effectively developing robot algorithms that appropriate and respond to those human signals. This could be as simple as picking up a cup of water based on a gesture or in more complex and important scenarios, aiding individuals with some kind of impairment to complete a task. It is for this reason that the eyes become extremely crucial here to not only determine the focus of an activity but to discern intent.

We normally use our eyes to interact with our immediate environment and express mental states and so the challenge presented to the robot is figuring out the focus as well as the intent of the human it is tasked with assisting.

In the same way that we see a myriad of different signals through the “eye gaze” (Who or what someone is listening to, agreement/disagreement, intrigue/interest, confusion or occupation with thought), we need machines to be able to do the same in order to effectively help us. These are signals that we as human beings also often have trouble discerning therefore in the context of robots assisting humans with tasks it becomes much more important for the focus and mental state to be discerned with speed and accuracy in order for it to be useful.

One way, A.I. can assist with the development of these robots is to remove subjectivity where possible. We saw this happen at the soccer world cup in 2022 where A.I. was used along with sensors and cameras to triangulate player positions for refereeing calls that are hard to make in real time. AI essentially converted a complicated job that relied on subjective perspectives and unflattering camera angles and turned it into a series of steps that takes input and provides you with an objective answer. The key takeaway here is AI was used to do away with the problem of subjective perspectives for objective calls and this is what we can hope to see with AI in robotics too.

It is possible we see objective perspectives that are more helpful (accurate) in discerning human intent than others because of factors such as the context of the task, how far along the machine is into completing it, any novel input it may have received (besides human expression) and any other information that it needs to reference before proceeding with the task. These are questions with objective answers and AI makes the process of arriving at conclusions which need to be used to discern human focus and intent much faster. Another area we can see an impact with AI is when the need arises to compute to complete tasks at speed; this is where we get to the subject of “Robotic Autonomy” which refers to the capability of a machine to be in an environment, understand itself, its tasks and complete them effectively.

This sometimes depends the robot needing to team with other robots or humans to complete a task effectively. One instance prevalent when human beings work together is when things need to move faster, we don’t always use language to convey intent. We see this in sports where it becomes important for players on a team to anticipate the intent and movement of their teammates in order to achieve a shared objective or outcome (goal or point). One way scientists and engineers attempt to replicate this in the design of robots through AI is by training the robot to absorb observations from the world (image, sensory input), compress it to the bare essential information it represents and then decompress it to what would be considered an acceptable human response to the original observation or input. There are a number of generative models that do this such as Probabilistic Graphical Models, Autoencoder Models, Generative Adversarial Networks and Flow-based Models. This type of training demands an understanding of the relationship between observed input (via cameras and sensors) and the essential features within it (Encoding) and then an accurate extraction of the data in the desired form. A simple example of this is with the text to text generation capability of GPT-4: An encoder would look over a sentence, “I like science” compress it into a code and translate it into a desired language.

For more information on the new AI program, visit the M.S in Artificial Intelligence Systems page.

For more information on the Cybersecurity & Network Administration program, visit the M.S in Cybersecurity & Network Administration page.