Trajectory generation device, trajectory generation method, and recording medium storing trajectory generation program
A trajectory generation device includes an acquisition unit, a clustering unit and a generation unit. The acquisition unit acquires successful trajectory information and failed trajectory information that are trajectory information representing sequences of states of a controlled object being taught by a teacher. The successful trajectory information is trajectory information when a task performed by the controlled object is successful, and the failed trajectory information is trajectory information when the task is failed. From the states of the controlled object belonging to the successful trajectory information and the states belonging to the failed trajectory information, the clustering unit generates clusters of successful classes of the states. On the basis of the clusters of successful classes, the generation unit generates target trajectory information representing a sequence of states of the controlled object, the sequence of states being usable as control targets when the controlled object is caused to execute the task.
1 . A trajectory generation device, comprising:
a processor configured to:
acquire trajectory information representing sequences of states of a controlled object being taught by a teacher, the trajectory information including successful trajectory information and failed trajectory information, the successful trajectory information being trajectory information when a task performed by the controlled object is successful, and the failed trajectory information being trajectory information when the task is failed;
from the states of the controlled object belonging to the successful trajectory information and the states of the controlled object belonging to the failed trajectory information, generate a cluster of a successful class of the states of the controlled object;
on the basis of the cluster of the successful class, generate target trajectory information representing a sequence of states of the controlled object, the sequence of states being used as control targets when the controlled object is caused to execute the task; and
render a display for providing guidance on a trajectory that is to be taught,
wherein the states of the controlled object are positions of the controlled object or positions and attitudes of the controlled object, the processor is configured to calculate a speed of the controlled object in each state, from the positions or the positions and attitudes included in the states of the controlled object belonging to the successful trajectory information and the failed trajectory information and the processor is configured to calculate degrees of similarity among the states of the controlled object belonging to the successful trajectory information and the failed trajectory information based on the positions or the positions and attitudes in the states and speeds in the states,
wherein the processor is configured specify an adjustment parameter to adjust the degrees of similarity so as to:
increase a degree of similarity between states of the controlled object that respectively belong to the successful trajectory information, and
reduce a degree of similarity between a state of the controlled object that belongs to the successful trajectory information and a state of the controlled object that belongs to the failed trajectory information.
2 . The trajectory generation device according to claim 1 , wherein the processor is configured to:
apply a clustering method that uses a Gaussian mixture model,
and
generate the cluster of the successful class on the basis of the calculated degrees of similarity.
3 . The trajectory generation device according to claim 2 , wherein the processor is configured to adjust the degrees of similarity among the states of the controlled object with the adjustment parameter.
4 . The trajectory generation device according to claim 2 , wherein the processor is configured to apply a Gaussian mixture regression method to the cluster of the successful class for generating the target trajectory information.
5 . A trajectory generation method, comprising a computer executing processing including:
acquiring trajectory information representing sequences of states of a controlled object being taught by a teacher, the trajectory information including successful trajectory information and failed trajectory information, the successful trajectory information being trajectory information when a task performed by the controlled object is successful, and the failed trajectory information being trajectory information when the task is failed;
from the states of the controlled object belonging to the successful trajectory information and the states of the controlled object belonging to the failed trajectory information, generating a cluster of a successful class of the states of the controlled object;
on the basis of the cluster of the successful class, generating target trajectory information representing a sequence of states of the controlled object, the sequence of states being used as control targets when the controlled object is caused to execute the task; and
rendering a display for providing guidance on a trajectory that is to be taught,
wherein the states of the controlled object are positions of the controlled object or positions and attitudes of the controlled object,
wherein generating a cluster of a successful class of the states of the controlled object comprises: calculating a speed of the controlled object in each state from the positions or the positions and attitudes included in the states of the controlled object belonging to the successful trajectory information and the failed trajectory information; and
calculating degrees of similarity among the states of the controlled object belonging to the successful trajectory information and the failed trajectory information based on the positions or the positions and attitudes in the states and speeds in the states,
wherein generating a cluster of a successful class of the states of the controlled object comprises:
specifying an adjustment parameter to adjust the degrees of similarity so as to:
increase a degree of similarity between states of the controlled object that respectively belong to the successful trajectory information, and
reduce a degree of similarity between a state of the controlled object that belongs to the successful trajectory information and a state of the controlled object that belongs to the failed trajectory information.
6 . The trajectory generation method according to claim 5 , wherein generating a cluster of a successful class of the states of the controlled object comprises:
applying a clustering method that uses a Gaussian mixture model; and
generating the cluster of the successful class on the basis of the calculated degrees of similarity.
7 . The trajectory generation method according to claim 5 , wherein generating target trajectory information comprises:
applying a Gaussian mixture regression method to the cluster of the successful class for generating the target trajectory information.
8 . The trajectory generation method according to claim 5 , wherein generating a cluster of a successful class of the states of the controlled object comprises:
adjusting the degrees of similarity among the states of the controlled object with the adjustment parameter.
9 . A non-transitory recording medium storing a trajectory generation program that is executable by a computer to perform processing, the processing comprising:
acquiring successful trajectory information relating to a successful trajectory when an operation by a controlled object is successful and failed trajectory information relating to a failed trajectory when an operation by the controlled object fails;
on the basis of the successful trajectory information and the failed trajectory information, clustering positions of the controlled object on the successful trajectory and positions of the controlled object on the failed trajectory into a successful class and a failed class by a clustering method specified in advance;
on the basis of the positions of the controlled object clustered in the successful class, generating target trajectory information relating to a target trajectory of the controlled object; and
rendering a display for providing guidance on a trajectory that is to be taught,
wherein clustering positions of the controlled object on the successful trajectory and positions of the controlled object on the failed trajectory into a successful class and a failed class by a clustering method specified in advance comprises:
calculating a speed of the controlled object from the positions of the controlled object on the successful trajectory and positions of the controlled object on the failed trajectory; and
calculating degrees of similarity among positions of the controlled object on the successful trajectory and positions of the controlled object on the failed trajectory based on the positions and speeds in the positions,
wherein clustering positions of the controlled object on the successful trajectory and positions of the controlled object on the failed trajectory into a successful class and a failed class by a clustering method specified in advance comprises:
specifying an adjustment parameter to adjust the degrees of similarity so as to:
increase a degree of similarity between positions of the controlled object that respectively belong to the successful trajectory information, and
reduce a degree of similarity between a position of the controlled object that belongs to the successful trajectory information and a position of the controlled object that belongs to the failed trajectory information.
10 . The non-transitory recording medium according to claim 9 , wherein clustering positions of the controlled object on the successful trajectory and positions of the controlled object on the failed trajectory into a successful class and a failed class by a clustering method specified in advance comprises:
applying the clustering method that uses a Gaussian mixture model; and
generating the cluster of the successful class on the basis of the calculated degrees of similarity.
11 . The non-transitory recording medium according to claim 9 , wherein clustering positions of the controlled object on the successful trajectory and positions of the controlled object on the failed trajectory into a successful class and a failed class by a clustering method specified in advance comprises:
adjusting the degrees of similarity among the positions of the controlled object with the adjustment parameter that adjusts the degrees of similarity.
12 . The non-transitory recording medium according to claim 9 , wherein generating target trajectory information comprises:
applying a Gaussian mixture regression method to the cluster of the successful class for generating the target trajectory information.