import numpy as np
from minestudio.simulator.callbacks.callback import MinecraftCallback
[docs]
class GateRewardsCallback(MinecraftCallback):
"""
A callback for calculating rewards based on the formation of a Nether portal.
This callback rewards the agent for building a valid Nether portal structure
using obsidian blocks.
"""
def __init__(self):
"""
Initializes the GateRewardsCallback.
"""
super().__init__()
self.prev_info = {}
self.reward_memory = {}
self.current_step = 0
[docs]
def reward_as_smlest_pos(self, obsidian_position, obsidian_positions):
"""
Calculates the reward for a potential portal frame based on a starting obsidian block.
It checks for both X-fixed and Z-fixed portal orientations.
:param obsidian_position: The (x, y, z) coordinates of a starting obsidian block.
:param obsidian_positions: A list of (x, y, z) coordinates of all obsidian blocks.
:return: The calculated reward for the best portal frame found from this starting block.
"""
x, y, z = obsidian_position
positive_pos = [(x, y, z), (x, y, z+1), (x, y, z+2), (x, y, z+3),
(x, y+1, z+3), (x, y+2, z+3), (x, y+3, z+3), (x, y+4, z+3),
(x, y+4, z+2), (x, y+4, z+1), (x, y+4, z),
(x, y+3, z), (x, y+2, z), (x, y+1, z)]
negtive_pos = [(x, y+1, z+1), (x, y+1, z+2), (x, y+2, z+2), (x, y+3, z+2), (x, y+3, z+1), (x, y+2, z+1)]
frame_num = len(set(positive_pos)&set(obsidian_positions))
extra_bonus = max(0, frame_num-12)
fix_x_reward = frame_num+extra_bonus - len(set(negtive_pos)&set(obsidian_positions)) - 0.1*len(set(obsidian_positions))
#fix z reward
positive_pos = [(x, y, z), (x+1, y, z), (x+2, y, z), (x+3, y, z),
(x+3, y+1, z), (x+3, y+2, z), (x+3, y+3, z), (x+3, y+4, z),
(x+2, y+4, z), (x+1, y+4, z), (x, y+4, z),
(x, y+3, z), (x, y+2, z), (x, y+1, z)]
negtive_pos = [(x+1, y+1, z), (x+2, y+1, z), (x+2, y+2, z), (x+2, y+3, z), (x+1, y+3, z), (x+1, y+2, z)]
frame_num = len(set(positive_pos)&set(obsidian_positions))
#extra_bonus = max(0, frame_num-8) + max(0, frame_num-10) + 2*max(0, frame_num-12) + 4*max(0, frame_num-14)
extra_bonus = max(0, frame_num-12)
fix_z_reward = frame_num+extra_bonus - len(set(negtive_pos)&set(obsidian_positions)) - 0.1*len(set(obsidian_positions))
larger_reward = max(fix_x_reward, fix_z_reward)
return larger_reward
[docs]
def gate_reward(self, info, obs = {}):
"""
Calculates the gate reward based on the current voxel information.
It iterates through all obsidian blocks and finds the maximum possible
portal reward.
:param info: The info dictionary containing voxel data.
:param obs: The observation dictionary (optional).
:return: The maximum gate reward.
"""
if "voxels" not in info:
return 0
voxels = info["voxels"]
obsidian_positions = []
for voxel in voxels:
if "obsidian" in voxel["type"]:
obsidian_positions.append((voxel["x"], voxel["y"], voxel["z"]))
max_reward = 0
for obsidian_position in obsidian_positions:
reward = self.reward_as_smlest_pos(obsidian_position, obsidian_positions)
max_reward = max(max_reward, reward)
return max_reward
[docs]
def after_reset(self, sim, obs, info):
"""
Resets the current step count and previous reward.
:param sim: The Minecraft simulator.
:param obs: The observation from the simulator.
:param info: Additional information from the simulator.
:return: The observation and info.
"""
self.current_step = 0
self.prev_reward = 0
return obs, info
[docs]
def after_step(self, sim, obs, reward, terminated, truncated, info):
"""
Calculates the gate reward for the current step.
The reward is the difference between the current gate reward and the previous
gate reward (delta reward).
:param sim: The Minecraft simulator.
:param obs: The observation from the simulator.
:param reward: The original reward from the simulator.
:param terminated: Whether the episode has terminated.
:param truncated: Whether the episode has been truncated.
:param info: Additional information from the simulator.
:return: The modified observation, overridden reward, terminated, truncated, and info.
"""
override_reward = 0.
cur_reward = self.gate_reward(info, obs)
override_reward = cur_reward - self.prev_reward
self.prev_reward = cur_reward
self.current_step += 1
return obs, override_reward, terminated, truncated, info
