Robots are playing an increasing role in the workplace, so the ability to work effectively with them is increasingly important. Research from Carnegie Mellon highlights how algorithms could be used to effectively allocate tasks to either human employees or robots.
The researchers reason that humans are better for some tasks, and robots for others, and effective understanding which is which is likely to be crucial to the smooth running of teams where both are present.
Act, learn or delegate
The researchers developed an algorithmic planner, which they refer to as “Act, Delegate or Learn (ADL). The planner considers a range of tasks before determining how best to assign them. The tool consists of three core questions: When should a robot act to complete a task? When should a task be delegated to a human? And when should a robot learn a new task?
“There are costs associated with the decisions made, such as the time it takes a human to complete a task or teach a robot to complete a task and the cost of a robot failing at a task,” the researchers explain. “Given all those costs, our system will give you the optimal division of labor.”
The team believes that their work could be invaluable in areas such as manufacturing where humans and robots collaborate on a wide range of tasks. The planner was put to the test on a relatively simple task where man and machine were required to insert blocks into a board and then stack parts of different shapes made out of Lego.
Smart allocation
The researchers accept that using algorithms to allocate tasks is certainly nothing new, and indeed platforms like Uber have deployed “algorithmic management” for a number of years. They believe, however, that their work is among the first to also include some kind of robot learning in their reasoning.
This helps to take into account the fact that robots are capable of learning new things and improving their capabilities. For instance, in manufacturing, robotic arms are often taught how to complete various tasks.
Such teaching can take a degree of time, so has an upfront cost associated with it. It can pay off in the long run, however, as the robot improves its capabilities. The challenge is often to understand when to take this time to teach the robot and when to simply allocate the task to a human. This assessment requires the system to accurately predict the kind of tasks the robot could perform once it has learned the new skill.
Once the planner is aware of this information it is able to convert the problem into what’s known as a mixed integer program, which is a form of optimization program that is often used in tasks like scheduling and planning. When put to the test, the planner was able to outperform existing models while also reducing the cost by around 15%.
