As robots become more common for both able-bodied individuals and those living with a disability, it is increasingly important that lay people be able to drive multi-degree-of-freedom platforms with low-dimensional controllers. One approach is to use state-conditioned action mapping methods to learn mappings between low-dimensional controllers and high DOF manipulators -- prior research suggests these mappings can simplify the teleoperation experience for users. Recent works suggest that neural networks predicting a local linear function are superior to the typical end-to-end multi-layer perceptrons because they allow users to more easily undo actions, providing more control over the system. However, local linear models assume actions exist on a linear subspace and may not capture nuanced actions in training data. We observe that the benefit of these mappings is being an odd function concerning user actions, and propose end-to-end nonlinear action maps which achieve this property. Unfortunately, our experiments show that such modifications offer minimal advantages over previous solutions. We find that nonlinear odd functions behave linearly for most of the control space, suggesting architecture structure improvements are not the primary factor in data-driven teleoperation. Our results suggest other avenues, such as data augmentation techniques and analysis of human behavior, are necessary for action maps to become practical in real-world applications, such as in assistive robotics to improve the quality of life of people living with w disability.

Artificial intelligence can be a powerful tool to help the United States military outthink its opponents, but some military officials are warning the U.S. is already falling behind the curve on implementing A.I. technology. Retired Rear Adm. Brian Losey, the former head of U.S. Naval Special Warfare Command, raised both the prospect of effectively using A.I. technology on the battlefield as well as concerns the U.S. has already fallen behind during a Tuesday panel discussion of the "The Promise and The Risk Of the A.I. Revolution" conference hosted by the U.S. Naval Institute, USNI News reported. "We're losing a lot of folks because of encounters with the unknown," Losey said. "Not knowing when we enter a house whether hostiles will be there and not really being able to adequately discern whether there's threats before we encounter them. And that's how we incurred most of our casualties."

In this interview, Audrow speaks with Andrea Bajcsy and Dylan Losey about a method that allows robots to infer a human's objective through physical interaction. They discuss their approach, the challenges of learning complex tasks, and on their experience collaborating between different universities. Some examples of people working with the more typical impedance control (left) and Bajcsy and Losey's learning method (right).

Researchers at Rice University have developed a way to train robots with just a little push. Their method uses algorithms that allow robots to not only respond to a human's touch in the moment, but alter their trajectory based on that physical input. "Here the robot has a plan, or desired trajectory, which describes how the robot thinks it should perform the task," said graduate student Dylan Losey about the project. "We introduced a real-time algorithm that modified, or deformed, the robot's future desired trajectory." Typically, when robots are programmed to respond to physical contact from a human, they usually only only do so in the moment and go right back to their original trajectory soon thereafter. But with the Rice team's algorithms, their robots were able to take that input and use it to adjust their trajectories in real time.