Files
Apollo3D_SE/behaviors/custom/reinforcement/walk/walk.py
2026-03-10 09:35:27 -04:00

147 lines
4.3 KiB
Python

import math
import os
import numpy as np
from behaviors.behavior import Behavior
from utils.math_ops import MathOps
from utils.neural_network import run_network, load_network
from scipy.spatial.transform import Rotation as R
class Walk(Behavior):
def __init__(self, agent):
super().__init__(agent)
self.joint_nominal_position = np.array(
[
0.0,
0.0,
0.0,
1.4,
0.0,
-0.4,
0.0,
-1.4,
0.0,
0.4,
0.0,
-0.4,
0.0,
0.0,
0.8,
-0.4,
0.0,
0.4,
0.0,
0.0,
-0.8,
0.4,
0.0,
]
)
self.train_sim_flip = np.array(
[
1.0,
-1.0,
1.0,
-1.0,
-1.0,
1.0,
-1.0,
-1.0,
1.0,
1.0,
1.0,
1.0,
-1.0,
-1.0,
1.0,
1.0,
-1.0,
-1.0,
-1.0,
-1.0,
-1.0,
-1.0,
-1.0,
]
)
self.scaling_factor = 0.5
self.previous_action = np.zeros(len(self.agent.robot.ROBOT_MOTORS))
self.model = load_network(model_path=os.path.join(os.path.dirname(__file__), "walk.onnx"))
def execute(self, reset: bool, target_2d: list, is_target_absolute: bool, orientation: float=None, is_orientation_absolute: bool=True) -> bool:
robot = self.agent.robot
world = self.agent.world
velocity = None
if is_target_absolute:
raw_target = target_2d - world.global_position[:2]
velocity = MathOps.rotate_2d_vec(raw_target, -robot.global_orientation_euler[2], is_rad=False)
else:
velocity = target_2d
rel_orientation = None
if orientation is None:
rel_orientation = MathOps.vector_angle(velocity) * 0.3
elif is_orientation_absolute:
rel_orientation = MathOps.normalize_deg(orientation - robot.global_orientation_euler[2])
else:
rel_orientation = orientation * 0.3
rel_orientation = np.clip(rel_orientation, -0.25, 0.25)
velocity = np.concatenate([velocity, np.array([rel_orientation])], axis=0)
velocity[0] = np.clip(velocity[0], -0.5, 0.5)
velocity[1] = np.clip(velocity[1], -0.25, 0.25)
radian_joint_positions = np.deg2rad(list(robot.motor_positions.values()))
radian_joint_speeds = np.deg2rad(list(robot.motor_speeds.values()))
qpos_qvel_previous_action = np.vstack(
(
[
(
(radian_joint_positions * self.train_sim_flip)
- self.joint_nominal_position
)
/ 4.6,
radian_joint_speeds / 110.0 * self.train_sim_flip,
self.previous_action / 10.0,
]
)
).T.flatten()
ang_vel = np.clip(np.deg2rad(robot.gyroscope) / 50.0, -1.0, 1.0)
orientation_quat_inv = R.from_quat(robot._global_cheat_orientation).inv()
projected_gravity = orientation_quat_inv.apply(np.array([0.0, 0.0, -1.0]))
#[0.5,0.25,0.25]
observation = np.concatenate(
[
qpos_qvel_previous_action,
ang_vel,
velocity,
projected_gravity,
]
)
observation = np.clip(observation, -10.0, 10.0)
nn_action = run_network(obs=observation, model=self.model)
target_joint_positions = (
self.joint_nominal_position + self.scaling_factor * nn_action
)
target_joint_positions *= self.train_sim_flip
self.previous_action = nn_action
for idx, target in enumerate(target_joint_positions):
robot.set_motor_target_position(
robot.ROBOT_MOTORS[idx], target*180/math.pi, kp=25, kd=0.6
)
def is_ready(self, *args, **kwargs) -> bool:
return True