JackM323/JackBot
1
0
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects 2 Releases Wiki Activity

simulation physics fixed

Browse Source 
fuck them physics
This commit is contained in:
JackM323
2026-08-06 15:18:06 +02:00
parent 523a4aea89
commit 346ec9e949
8 changed files with 457 additions and 900 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ml/run_eval_training.py
+6 -13
View File
@@ -6,31 +6,27 @@ import numpy as np
from ml.env import JackBotEnv, CurriculumPhase
def evaluate_kinematics(episode_length: int = 1000):
# Use kinematics_only mode so inverse kinematics generates gait motion from commands
env = JackBotEnv(
use_gui=True,
random_command=False,
max_episode_steps=episode_length,
robot_mode="kinematics_only"
robot_mode="kinematics"
)
print("\n" + "=" * 70)
print(" RUNNING MULTI-PHASE REWARD BENCHMARK (KINEMATICS MODE)")
print("=" * 70 + "\n")
# Define test suite covering every curriculum stage
phase_configs = [
(CurriculumPhase.STAND_ONLY, "STAND ONLY", np.array([0.0, 0.0, 0.0, 0.0], dtype=np.float32)),
(CurriculumPhase.FORWARD, "FORWARD GAIT", np.array([0.1, 0.0, 0.0, 0.0], dtype=np.float32)),
(CurriculumPhase.TURN_AND_DIRECTION, "FORWARD + YAW TURN", np.array([0.3, 0.0, 0.0, 0.4], dtype=np.float32)),
(CurriculumPhase.OMNI_DIRECTION, "STRIDE LATERAL", np.array([0.2, 0.3, 0.0, 0.0], dtype=np.float32)),
(CurriculumPhase.FULL_COMMAND, "FULL OMNI COMBINATION",np.array([0.3, 0.2, 0.0, 0.3], dtype=np.float32)),
(CurriculumPhase.FORWARD, "FORWARD GAIT", np.array([1.0, 0.0, 0.0], dtype=np.float32)),
(CurriculumPhase.TURN_AND_DIRECTION, "FORWARD + YAW TURN", np.array([0.5, 0.0, 0.4], dtype=np.float32)),
(CurriculumPhase.OMNI_DIRECTION, "STRIDE LATERAL", np.array([0.5, 0.5, 0.0], dtype=np.float32)),
(CurriculumPhase.FULL_COMMAND, "FULL OMNI COMBINATION",np.array([0.3, 0.2, 0.3], dtype=np.float32)),
]
for phase_enum, label, cmd in phase_configs:
obs, _ = env.reset()
# Force specific curriculum phase & target command
env.curriculum_phase = phase_enum
env.command = cmd.copy()
@@ -39,9 +35,7 @@ def evaluate_kinematics(episode_length: int = 1000):
step_count = 0
while not done:
# Action array is unused in kinematics_only mode
dummy_action = np.zeros(18, dtype=np.float32)
obs, reward, terminated, truncated, _ = env.step(dummy_action)
total_reward += reward
step_count += 1
@@ -49,11 +43,10 @@ def evaluate_kinematics(episode_length: int = 1000):
time.sleep(1.0 / 60.0)
# Retrieve detailed component averages
comp_averages = env.get_reward_component_averages()
print(f"\n--- Episode Stage: [{phase_enum.name}] ({label}) ---")
print(f"Command Applied: vx={cmd[0]:.2f}, vy={cmd[1]:.2f}, vz={cmd[2]:.2f}, yaw={cmd[3]:.2f}")
print(f"Command Applied: vx={cmd[0]:.2f}, vy={cmd[1]:.2f}, yaw={cmd[2]:.2f}")
print(f"Total Episode Reward: {total_reward:.4f}")
print("Component Step Averages:")
for name, value in comp_averages.items():