ADAMANT: Adaptive Manipulation for Tasks

Robots will play an important role in NASA#39;s upcoming missions to the Moon and beyond.nbsp; They will need to manipulate their environment in complex and useful ways - carrying objects, using tools, and assisting crew.nbsp; NASArsquo;s humanoid robots have highly dexterous end-effectors, but developing software to fully utilize such hands remains a challenging task.nbsp; Grasping strategies are highly dependent on object models and localization.nbsp; Environmental obstacles or the object#39;s intended use can strongly influence how best to grasp it.nbsp;Previously with NASA, TRACLabs developed CRAFTSMAN, which supports robot-independent task descriptions, although grasp strategies are robot-specific.nbsp; Here, we extend CRAFTSMAN to handle grasping as a task-informed behavior.nbsp; This new system, called ADAMANT, will connect to other CRAFTSMAN software nodes to help find the best option for acquiring an object.nbsp; The result will be a robot grasping interface that produces more robust robot behaviors while reducing the cognitive load on remote robot operators.The ADAMANT system uses sensor and/or model data in addition to a task description to develop a ranked list of potential grasps for an object, using user-selected grasp metrics.nbsp; These different grasps are explored in the context of the complete task to arrive at the strategy most likely to succeed. In Phase I, we demonstrated that we could describe tasks in terms of the effect on the object, rather than just a sequence of waypoints for a manipulator and gripper.nbsp; In Phase II, we will fully incorporate this object-centric idea into CRAFTSMAN, allowing the user to define tasks without a specific manipulator/gripper in mind.nbsp; The ADAMANT system will automatically figure out at run-time the best way to grasp an object given models of the hand and models (or sensor data) of the object.This work will make it easier for NASA to use robots in conjunction with pre-existing operational procedures, and has many applications to industrial robotics.