def configure(self, config):
self.set_common_parameters(config)
self.planning_depth = config.model_predictive_rl.planning_depth
self.do_action_clip = config.model_predictive_rl.do_action_clip
if hasattr(config.model_predictive_rl, 'sparse_search'):
self.sparse_search = config.model_predictive_rl.sparse_search
self.planning_width = config.model_predictive_rl.planning_width
self.share_graph_model = config.model_predictive_rl.share_graph_model
self.linear_state_predictor = config.model_predictive_rl.linear_state_predictor
if self.linear_state_predictor:
self.state_predictor = LinearStatePredictor(config, self.time_step)
graph_model = RGL(config, self.robot_state_dim, self.human_state_dim)
self.value_estimator = ValueEstimator(config, graph_model)
self.model = [graph_model, self.value_estimator.value_network]
else:
if self.share_graph_model:
graph_model = RGL(config, self.robot_state_dim, self.human_state_dim)
self.value_estimator = ValueEstimator(config, graph_model)
self.state_predictor = StatePredictor(config, graph_model, self.time_step)
self.model = [graph_model, self.value_estimator.value_network, self.state_predictor.human_motion_predictor]
else:
graph_model1 = RGL(config, self.robot_state_dim, self.human_state_dim)
self.value_estimator = ValueEstimator(config, graph_model1)
graph_model2 = RGL(config, self.robot_state_dim, self.human_state_dim)
self.state_predictor = StatePredictor(config, graph_model2, self.time_step)
self.model = [graph_model1, graph_model2, self.value_estimator.value_network,
self.state_predictor.human_motion_predictor]
logging.info('Planning depth: {}'.format(self.planning_depth))
logging.info('Planning width: {}'.format(self.planning_width))
logging.info('Sparse search: {}'.format(self.sparse_search))
if self.planning_depth > 1 and not self.do_action_clip:
logging.warning('Performing d-step planning without action space clipping!')
def __init__(self, config, robot_state_dim, human_state_dim, shared_gcn=True):
super().__init__()
self.shared_gcn = shared_gcn
if shared_gcn:
self.graph_model = RGL(config, robot_state_dim, human_state_dim)
else:
self.graph_model_val = RGL(config, robot_state_dim, human_state_dim)
self.graph_model_act = RGL(config, robot_state_dim, human_state_dim)
self.value_network = mlp(config.gcn.X_dim, config.rgl_ppo.value_network_dims)
self.action_network = mlp(config.gcn.X_dim, config.rgl_ppo.value_network_dims[:-1] +
[4])