refined primitive skills (178907d5) · Commits · 张泽凯 / VAGEN

vagen/env/init.py

+6 −1

Original line number	Diff line number	Diff line
		@@ -2,7 +2,7 @@ from .sokoban import SokobanEnv,SokobanEnvConfig
		from .frozenlake import FrozenLakeEnv,FrozenLakeEnvConfig, FrozenLakeService
		from .navigation import NavigationEnv, NavigationEnvConfig, NavigationServiceConfig, NavigationService
		from .svg import SVGEnv, SvgEnvConfig, SVGService

		from .primitive_skill import PrimitiveSkillEnv, PrimitiveSkillEnvConfig, PrimitiveSkillService
		REGISTERED_ENV = {
		"sokoban": {
		"env_cls": SokobanEnv,
		@@ -24,4 +24,9 @@ REGISTERED_ENV = {
		"config_cls": SvgEnvConfig,
		"service_cls": SVGService
		},
		"primitive_skill": {
		"env_cls": PrimitiveSkillEnv,
		"config_cls": PrimitiveSkillEnvConfig,
		"service_cls": PrimitiveSkillService
		}
		}
		No newline at end of file

vagen/env/navigation/env.py

+1 −1

Original line number	Diff line number	Diff line
		@@ -347,7 +347,7 @@ class NavigationEnv(BaseEnv):
		Returns:
		System prompt string
		"""
		if self.config.visual_env:
		if self.config.render_mode == "vision":
		return system_prompt_vision.format(
		max_actions_per_step=self.config.max_actions_per_step,
		action_sep=self.config.action_sep

vagen/env/navigation/env_config.py

+2 −2

Original line number	Diff line number	Diff line
		@@ -10,7 +10,7 @@ class NavigationEnvConfig(BaseEnvConfig):
		down_sample_ratio: float = 1.0
		fov: int = 100
		multiview: bool = False
		visual_env: bool = True
		render_mode: str= 'vision'
		max_actions_per_step: int = 10
		max_action_penalty: float = -0.1
		format_reward: float = 0.5
		@@ -19,7 +19,7 @@ class NavigationEnvConfig(BaseEnvConfig):
		def config_id(self) -> str:
		"""Generate a unique identifier for this configuration."""
		id_fields = ["resolution", "eval_set", "exp_name", "down_sample_ratio",
		"fov", "multiview", "visual_env", "max_actions_per_step"]
		"fov", "multiview", "render_mode", "max_actions_per_step"]
		id_str = ",".join([f"{field.name}={getattr(self, field.name)}" for field in fields(self) if field.name in id_fields])
		return f"NavigationEnvConfig({id_str})"

vagen/env/primitive_skill/init.py

0 → 100644

+3 −0

Original line number	Diff line number	Diff line
		from .env import PrimitiveSkillEnv
		from .env_config import PrimitiveSkillEnvConfig
		from .service import PrimitiveSkillService
		No newline at end of file

vagen/env/primitive_skill/env.py

+12 −6

Original line number	Diff line number	Diff line
		@@ -12,14 +12,18 @@ import vagen.env.primitive_skill.maniskill.env
		class PrimitiveSkillEnv(BaseEnv):
		def __init__(self, config: PrimitiveSkillEnvConfig):
		self.config = config
		self.env=build_env(config.env_id,record_dir='./test')
		if self.config.record_video:
		record_dir = self.config.video_record_dir
		else:
		record_dir = None
		self.env=build_env(config.env_id,record_dir=record_dir)

		def reset(self, seed: Optional[int] = None) -> Tuple[Dict[str, Any], Dict[str, Any]]:
		self.total_reward = 0
		_, info=self.env.reset(seed=seed)
		obs=self._render(info,init_obs=True)
		self.last_info=info
		return obs, info
		return obs, {}

		def step(self,action_str):
		reward=0
		@@ -34,12 +38,14 @@ class PrimitiveSkillEnv(BaseEnv):
		valid_actions = []
		metrics = {
		"turn_metrics": {
		"action_is_valid": True, # True if at least one valid action was parsed
		"action_is_valid": False, # True if at least one valid action was parsed
		},
		"traj_metrics": {
		"success": False, # Will be set to True if agent reaches goal
		},
		}
		info=self.last_info
		terminated, truncated = False, False
		for action in rst['actions']:
		parsed_action = self._parse_action(action)
		if parsed_action is not None:
		@@ -49,10 +55,10 @@ class PrimitiveSkillEnv(BaseEnv):
		else:
		info=self.last_info
		terminated, truncated = False, False
		metrics["turn_metrics"]['action_is_valid'] = False
		break
		if truncated or terminated:
		break
		metrics["turn_metrics"]['action_is_valid'] = len(valid_actions) > 0 and len(valid_actions)==len(rst['actions'])
		if metrics["turn_metrics"]['action_is_valid']:
		reward += self.config.format_reward
		if info['is_success']:
		@@ -162,7 +168,7 @@ class PrimitiveSkillEnv(BaseEnv):
		action_array[2] = 1
		else:
		# Invalid action name
		return np.zeros(9)
		return None

		# Extract parameters
		params_str = action_str.split('(')[1].split(')')[0]
		@@ -204,7 +210,7 @@ if __name__ == "__main__":
		This code demonstrates how to create an instance of the environment,
		reset it, and interact with it using manual input actions.
		"""
		config = PrimitiveSkillEnvConfig()
		config = PrimitiveSkillEnvConfig(record_video=True, video_record_dir="./test_manipulation_video")
		env = PrimitiveSkillEnv(config)

		print(env.system_prompt())

Admin message