The slow and deliberate approach of the arcade claw is similar to state-of-the-art pick-and-place robots, which use high-level planners to process visual images and plan out a series of moves to grab an object. If a gripper misses its mark, it’s back to the starting point, where the controller must map out a new plan.

MIT engineers have now developed a robot gripper that grasps by reflex. Instead of starting from scratch after a failed attempt, the robot adapts in the moment to reflexively roll, palm, or pinch an object to get a better hold. It can make these adjustments without engaging a higher-level planner, just like how a person might fumble in the dark for a bedside glass without much conscious thought.

This new design is the first to incorporate reflexes into a robotic planning architecture. While it’s currently a proof of concept, it provides a structure for embedding reflexes into a robotic system. The researchers plan to program more complex reflexes in the future to create nimble, adaptable machines that can work with and among humans in ever-changing settings.

“In environments where people live and work, there’s always uncertainty,” says Andrew SaLoutos, a graduate student in MIT’s Department of Mechanical Engineering. “We’re hoping a robot with reflexes could adapt and work with this kind of uncertainty.”

Presenting their work at the IEEE International Conference on Robotics and Automation (ICRA) in May, SaLoutos and his colleagues aim to create a more human-like touch for robots by incorporating reflexive capabilities into their grippers.

High and low

Traditional robotic grippers are designed for slow and precise tasks, such as repetitive assembly line work in factories. These systems rely on visual data from onboard cameras, which limits their reaction time when trying to recover from a failed grasp.

“There’s no way to react quickly in those situations,” SaLoutos explains. “Their only option is to start over, which takes a lot of time.”

In their new work, the MIT team built a more reactive platform using fast actuators originally developed for their nimble, four-legged robot, the mini cheetah.

The team’s design features a high-speed arm and two lightweight fingers with custom sensors that instantly record force, location, and proximity of contacts more than 200 times per second.

The researchers designed the robotic system to initially process visual data of a scene using a high-level planner. The planner marks the object’s current location for picking it up and where the robot should place it down. Then, a path for the arm to reach and grasp the object is set. At this point, the reflexive controller takes over.

If the gripper fails to grab hold of the object, instead of starting over, the robot carries out one of three grasp maneuvers called “reflexes.” These reflexes are triggered by real-time measurements at the fingertips and allow the fingers to grab, pinch, or drag the object for a better hold. The reflexes are carried out without involving the high-level planner, behaving almost instinctively without carefully evaluating the situation for an optimal fix.

According to Sangbae Kim, a professor of mechanical engineering and director of the Biomimetic Robotics Laboratory at MIT, this structure is like building a trust system and delegating tasks to lower-level divisions instead of having the CEO micromanage everything. Waiting for the optimal solution can often make the situation worse or irrecoverable.

Cleaning via reflex

The team demonstrated the gripper’s reflexes by clearing a cluttered shelf. They placed various household objects on a shelf and showed that the robot was able to quickly adapt its grasp to each object’s shape. Out of 117 attempts, the gripper successfully picked and placed objects over 90% of the time, without starting over after a failed grasp.

In a second experiment, researchers shifted the position of a cup and the gripper, even without a visual update, readjusted and sensed the cup in its grasp. Compared to a baseline grasping controller, the gripper’s reflexes increased successful grasps by over 55%.

The engineers are now working on including more complex reflexes and grasp maneuvers in the system to build a general pick-and-place robot that can adapt to cluttered and constantly changing spaces.

“Picking up a cup from a clean table was solved 30 years ago,” Kim says. “But a more general approach, like picking up toys in a toybox or a book from a library shelf, has not been solved. Now with reflexes, we think we can one day pick and place in every possible way, so that a robot could potentially clean up the house.”

Support for this research was provided by the Advanced Robotics Lab of LG Electronics and the Toyota Research Institute.

Video: https://youtu.be/XxDi-HEpXn4