File size: 1,474 Bytes
d94c87e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 | """Faithful minimal stand-ins for the openpi transforms used by g1_dex3_policy."""
import dataclasses
from typing import Sequence
import numpy as np
class DataTransformFn:
pass
def pad_to_dim(x, target_dim, axis=-1):
current = x.shape[axis]
if current >= target_dim: return x
pad = [(0,0)]*x.ndim
pad[axis]=(0, target_dim-current)
return np.pad(x, pad)
def make_bool_mask(*dims):
out=[]
for d in dims:
out.extend([True]*d if d>0 else [False]*(-d))
return tuple(out)
@dataclasses.dataclass(frozen=True)
class DeltaActions(DataTransformFn):
mask: Sequence[bool] | None
def __call__(self, data):
if "actions" not in data or self.mask is None: return data
state, actions = data["state"], np.array(data["actions"])
mask = np.asarray(self.mask); dims = mask.shape[-1]
actions[..., :dims] = np.where(mask, actions[..., :dims]-state[..., :dims], actions[..., :dims])
data["actions"]=actions; return data
@dataclasses.dataclass(frozen=True)
class AbsoluteActions(DataTransformFn):
mask: Sequence[bool] | None
def __call__(self, data):
if "actions" not in data or self.mask is None: return data
state, actions = data["state"], np.array(data["actions"])
mask = np.asarray(self.mask); dims = mask.shape[-1]
actions[..., :dims] = np.where(mask, actions[..., :dims]+state[..., :dims], actions[..., :dims])
data["actions"]=actions; return data
|