Honda Motor Co., Ltd.

Chyon Hae Kim

Hiroshi Tsujino

Shigeki Sugano

Hiroyuki Nakahara

Shota Yamazaki

A path planning system for bringing state of an object into a target state includes a search tree production unit for producing in advance, in a state space with said target state defined as a root, a search tree having a branch at each one of a plurality of sections of the state space, said state space being divided into the plurality of sections in advance. The system also includes a search tree memory unit for storing the search tree, and a path generation unit for determining, a route on the search tree from the branch corresponding to the current state to the root. The path planning/control system further includes a path control unit for controlling the path of the object to bring the state of the object into the target state in accordance with the route on the search tree determined by the path planning system.

Method and system for path planning and controlling

A trajectory planning method according to the present invention is one for obtaining a trajectory for controlling a state of an object toward a goal state by a trajectory planning system. The method includes the steps of dividing, by a cell generating section for dividing a state space of the object into cells, the state space into cells in such a way that approximation error due to discretization is minimized for a predetermined number of cells; generating, by a search tree generating section, a search tree which corresponds to state transition of the object in such a way that each cell does not contain more than one of nodes of branches of the search tree; and determining, by a trajectory generating section, a path from the current state to the goal state of the object using the search tree.

Trajectory planning method, trajectory planning system and trajectory planning and control system

A trajectory planning system obtains a trajectory for controlling a state of an object toward a goal state. The system includes a search tree generating section which registers a state of the object as a root of a search tree in a state space, registers a next state of the object after a lapse of a predetermined time interval obtained through dynamical relationships during the time interval as a branch of the search tree in the state space. The system further includes a known-state registration tree storing section which stores a known-state registration tree and a known-state registration tree generating section which determines a cell to which the next state belongs among a plurality of cells previously prepared by segmenting the state space, determines whether or not a state which belongs to the cell has already been registered as a branch of the known-state registration tree, discards the next state when a state which belongs to the cell has been registered, and registers the next step as a branch of the known-state registration tree when a state which belongs to the cell has not been registered. The system further includes a trajectory generating section which selects a state whose distance to the goal state is minimum among states registered as branches of the known-state registration tree and obtains a trajectory using a sequence of states in a backward direction from the state toward the root of the known-state registration tree.

Trajectory planning method, trajectory planning system and robot

A learning control system according to the present invention is one which performs learning of action values of actions in an apparatus which identifies its state as one of predetermined states, and selects an action based on the obtained action values and the identified state. The learning control system includes n action value learning devices including the first to the n th learning devices which perform learning of n action values from Qto Q, assuming that n is a positive integer and an action value determining device which determines the total action value of an action Q of each state based on outputs of the n action value learning devices. In the learning control system, the first target value of the first action value learning device is determined based on the reward r obtained after an action has been carried out by the next state and a total action value Q' that was prepared for the action selection in the next state, and the first learning device updates the first action value Qusing the first target value. When n is 2 or more, the n-th a target value of the n th action value learning device is set to the difference between the (n&#x2212;1) th target value of the (n&#x2212;1) th learning device and the action value Q, and the n th learning device updates the n th action value Qusing the n th target value.

Growing community of inventors

Wako, Japan

Chyon Hae Kim