Robot controller
Motions of a robot are defined by a plurality of frames P 0 to P 11 at a plurality of different time points. The frames include a plurality of reference frames. In the reference frames, the robot stands alone without falling. The data of the other frames, i.e., frames other than the reference frames, are set only roughly before the robot begins walking. When the robot starts walking, the roughly set data of the other frames is corrected based on control information calculated from the roughly set data and the data of the reference frames.
1. A robot controller that controls motions of a robot, comprising:
a posture information acquiring unit that acquires posture information indicative of a posture of the robot at each of a plurality of different time points, the postures being generated before the robot starts walking and including at least one reference posture in which the robot stands alone without falling down; and
a motion controlling unit that controls motions of the robot in between any two of the postures by using a recurrent neural network (RNN) circuit, wherein
control information is calculated based on the posture information of at least the two postures such that the motion of the robot reaches the reference posture.
2. The robot controller according to claim 1 , wherein the robot is a two-legged robot, and the reference posture is a posture in which the robot stands alone only on one leg and a stride is zero.
3. The robot controller according to claim 1 , further comprising:
a motion state information acquiring unit that acquires motion information indicative of a state of motions of the robot; and
a correcting unit that corrects the posture information acquired by the posture information acquiring unit based on the motion information acquired by the motion state information acquiring unit, wherein
the motion controlling unit calculates the control information based on corrected posture information.
4. The robot controller according to claim 1 , wherein the robot is a two-legged robot, and the robot controller further comprising:
a reaction force information acquiring unit that acquires force information indicative of a reaction force received by a sole of each of the legs of the robot from a floor surface on which the robot walks; and
a gain adjusting unit that adjust a gain of the force information based on a state of the leg of the robot, wherein
the motion controlling unit calculates the control information based on the gain adjusted by the gain adjusting unit.
5. The robot controller according to claim 1 , further comprising a request receiving unit that receives a request for changing motion of the robot, wherein
the motion controlling unit provides a control to change motion of the robot based on the request received by the request receiving unit.
6. The robot controller according to claim 1 , wherein the recurrent neural network (RNN) circuit controls the motions of the robot using a piecewise linear function modulated by a decaying exponential.
7. The robot controller according to claim 6 , wherein the recurrent neural network (RNN) circuit controls the motions of the robot using a combination of basic recurrent neural network circuits generating the piecewise linear function, and signal output from the combination of the basic recurrent neural network circuits is smoothed by the decaying exponential that functions as a low-pass filter before the signal is sent to a motor for controlling movements of a joint.
8. The robot controller according to claim 1 , wherein the recurrent neural network (RNN) circuit controls the motions of the robot using a combination of basic recurrent neural network circuits generating piecewise linear functions, and signal output from the combination of the basic recurrent neural network circuits is smoothed by a low-pass filter before the signal is sent to a motor for controlling movements of a joint.
9. A robot controller that controls motions of a two-legged robot, comprising:
a posture information acquiring unit that acquires posture information indicative of a posture of the robot at each of a plurality of different time points, the postures including at least one reference posture in which the robot stands alone without falling down;
a motion controlling unit that controls motions of the robot in between any two of the postures by calculating control information based on the posture information of at least the two postures such that the motion of the robot reaches the reference posture;
a reaction force information acquiring unit that acquires force information indicative of a reaction force received by a sole of each of the legs of the robot from a floor surface on which the robot walks; and
a gain adjusting unit that adjusts a gain of the force information based on a state of the leg of the robot, wherein
the motion controlling unit calculates the control information based on the gain adjusted by the gain adjusting unit, and the gain adjusting unit sets, when the leg of the robot lands on the floor surface, a gain of the force information smaller than a gain before the leg of the robot lands on the floor surface.
10. A robot controller that controls motions of a two-legged robot, comprising:
a posture information acquiring unit that acquires posture information indicative of a posture of the robot at each of a plurality of different time points, the postures including at least one reference posture in which the robot stands alone without falling down;
a motion controlling unit that controls motions of the robot in between any two of the postures by calculating control information based on the posture information of at least the two postures such that the motion of the robot reaches the reference posture;
a reaction force information acquiring unit that acquires force information indicative of a reaction force received by a sole of each of the legs of the robot from a floor surface on which the robot walks; and
a gain adjusting unit that adjusts a gain of the force information based on a state of the leg of the robot, wherein
the motion controlling unit calculates information based on the gain adjusted by the gain adjusting unit, and the gain adjusting unit sets, when one leg of the robot lands on the floor surface and other leg is thrown forward of the robot, a gain of the force information smaller than a gain before the other leg is thrown forward.
11. A method of controlling motions of a robot, comprising:
a posture information acquiring step of acquiring posture information indicative of a posture of the robot at each of a plurality of different time points, the postures being generated before the robot starts walking and including at least one reference posture in which the robot stands alone without falling down; and
a motion controlling step of controlling motions of the robot in between any two of the postures by using a recurrent neural network (RNN) circuit, wherein
control information is calculated based on the posture information of at least the two postures such that the motion of the robot reaches the reference posture.
12. A robot controller that controls motions of a robot, comprising:
a posture information acquiring unit that acquires posture information indicative of a posture of the robot at each of a plurality of different time points, the postures being previously generated and including at least one reference posture in which the robot stands alone without falling down; and
a motion controlling unit that controls motions of the robot in between any two of the postures by using a recurrent neural network (RNN) circuit, wherein
control information is calculated based on the posture information of at least the two postures such that the motion of the robot reaches the reference posture,
the recurrent neural network (RNN) circuit controls the motions of the robot using a combination of basic recurrent neural network circuits generating piecewise linear functions, and signal output from the combination of the basic recurrent neural network (RNN) circuits is smoothed by a low-pass filter before the signal is sent to a motor for controlling movements of a joint.