''' This script processes `office_changes.json` to generate a dataset of office manipulation tasks. The `CHANGES_DESCRIPTION_FILE` file contains a list of tasks, each with: - "name": A natural language description. - "init_conditions": An array of conditions to define the starting world state. null if no changes - "goal_conditions": An array of conditions that describe the target world state. - "actions": A sequence of actions to transition from the initial to the goal state. ''' import json import pickle from office_graph import office_graph, add_edges import copy CHANGES_DESCRIPTION_FILE = 'office_changes.json' OUT_FILENAME = 'office29.pkl' def change_node(node, condition, graph): if condition.get("holding", False): graph.nodes[node]['holding'] = (condition["holding"]['name'], condition["holding"]['id']) if condition["relation"]: graph.nodes[node]['related_to'] = (condition["related_to"]["name"], condition["related_to"]["id"]) graph.nodes[node]['relation'] = condition["relation"] if condition["states"]: states = condition.get("states", []) for state in states: if state not in graph.nodes[node]['states']: graph.nodes[node]['states'].append(state) if state == 'off': graph.nodes[node]['states'].remove('on') if state == 'on': graph.nodes[node]['states'].remove('off') if state == 'closed': graph.nodes[node]['states'].remove('open') if state == 'open': graph.nodes[node]['states'].remove('closed') if __name__ == "__main__": # Load the JSON file with open(CHANGES_DESCRIPTION_FILE, "r") as file: tasks = json.load(file) dataset = [] i=0 for task in tasks: i+=1 name = task["name"] init_graph = copy.deepcopy(office_graph) goal_graph = copy.deepcopy(office_graph) # Process initial conditions if task['init_conditions']: for condition in task.get("init_conditions", []): node = (condition["node"]["name"], condition["node"]["id"]) change_node(node, condition, init_graph) # Process goal conditions for condition in task.get("goal_conditions", []): node = (condition["node"]["name"], condition["node"]["id"]) change_node(node, condition, goal_graph) # Process actions actions = [] for action in task.get("actions", []): actions.append((action['action'], (action['node']["name"],action['node']["id"]))) # Reset graph edges init_graph.remove_edges_from(list(office_graph.edges)) goal_graph.remove_edges_from(list(office_graph.edges)) add_edges(init_graph) add_edges(goal_graph) entity = { "id": i, "name": name, "task": name, "init": init_graph, "goal": goal_graph, "actions": actions, } dataset.append(entity) with open(OUT_FILENAME, "wb") as file: pickle.dump(dataset, file)