SelfRising/RL/policy.py

17 lines
796 B
Python

"""Standalone NumPy speed-command policy; no simulated state or critic access."""
from pathlib import Path
import numpy as np
class Policy:
def __init__(self,path):
with np.load(Path(path),allow_pickle=False) as d:self.layers=[(d[f'w{i}'].copy(),d[f'b{i}'].copy()) for i in range(3)]
for (w,b),shape in zip(self.layers,[(64,196),(64,64),(2,64)]):
if w.shape!=shape or b.shape!=(shape[0],) or not np.isfinite(w).all() or not np.isfinite(b).all():raise ValueError('Invalid speed actor')
def reset(self):pass
def __call__(self,obs):
x=np.asarray(obs,dtype=np.float32)
if x.shape!=(196,) or not np.isfinite(x).all():raise ValueError('Expected measured IMU/command history196')
for w,b in self.layers:x=np.tanh(w@x+b)
return x