Spaces:
Running on Zero
Running on Zero
File size: 3,527 Bytes
e95ffc8 | 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 42 43 44 45 46 47 48 49 50 51 52 53 54 55 | """Manufacturing-MoE current robotics: inference with observable pre-target history."""
import argparse,json,sys
from pathlib import Path
import numpy as np,torch
from robotics_calibrated import RoboticsMoECalibrated,PreparedCalibrated,prediction
ROOT=Path(__file__).resolve().parents[1]
sys.path.insert(0,str(ROOT/'runtime'))
from robot_physics import simulate
from robot_features import public_features
from gravity_features import signed_features
PRIVATE_DEFAULTS={'private_motor_efficiency':1.,'private_driver_loss':1.,'private_friction_scale':1.,'private_settle_s':0.,'private_regen_scale':1.}
def robot_predict(checkpoint,robot,motion,history_probes=None,past_calibration=None):
if history_probes is None:history_probes=robot.get('history_probes')
if past_calibration is None:past_calibration=robot.get('past_calibration')
if history_probes is None or past_calibration is None:
raise ValueError('Calibrated robotics model requires seven previous power/settle observations and six independently measured calibration cycles (24 values).')
h=np.asarray(history_probes,np.float32)
hc=np.asarray(past_calibration,np.float32)
if h.shape!=(7,) or hc.shape!=(24,) or not np.isfinite(h).all() or not np.isfinite(hc).all():
raise ValueError('History observations must be finite and of length 7 and 24, respectively.')
state=torch.load(checkpoint,map_location='cpu',weights_only=False)
s=state.get('robotics_bundle',state)
if s['model_type']!='RoboticsMoECalibrated':raise ValueError('Checkpoint does not contain calibrated robotics model.')
model=RoboticsMoECalibrated();model.load_state_dict(s['robotics']);model.eval()
device='cuda' if torch.cuda.is_available() else 'cpu'
model.to(device)
cfg={**robot,**PRIVATE_DEFAULTS}
out=simulate(cfg,motion,steps=22)
joint,mask,global_x=public_features(cfg,motion,out)
signed_joint,signed_global=signed_features(cfg,motion)
d={'joint':joint[None],'mask':mask[None],'global_features':global_x[None],
'joint_gravity':signed_joint[None],'global_gravity':signed_global[None],
'history':h[None],'history_calibration':hc[None],
'prior':np.array([[out['nominal_time_s'],out['nominal_energy_Wh']]],np.float32),
'y':np.ones((1,2),np.float32),'n_axes':np.array([robot['n']]),
'topology':np.array([robot['topology']]),'robot_ids':np.array([robot.get('robot_id','inference')])}
scaler=s['train_normalizer'];ds=PreparedCalibrated(d,scaler['mean'],scaler['std'])
pred=prediction(model,ds,device=device)[0]
return {'model_family':'robotics','joint_count':robot['n'],'topology':robot['topology'],
'predicted_cycle_time_s':float(pred[0]),'predicted_electricity_Wh':float(pred[1]),
'material_removed_g':0.,'history_calibration_cycles':6,
'signed_gravity_features':True,'simulator_nominal_energy_Wh':out['nominal_energy_Wh'],
'not_a_real_machine_validated_model':True}
if __name__=='__main__':
pa=argparse.ArgumentParser()
pa.add_argument('--checkpoint',default='reports/v51/model.pt');pa.add_argument('--robot',required=True)
pa.add_argument('--motion',required=True);pa.add_argument('--output',default='outputs/robotics_v51_prediction.json')
a=pa.parse_args()
r=json.loads(Path(a.robot).read_text());mo=json.loads(Path(a.motion).read_text())
pred=robot_predict(a.checkpoint,r,mo)
o=Path(a.output);o.parent.mkdir(parents=True,exist_ok=True);o.write_text(json.dumps(pred,indent=2))
print(json.dumps(pred,indent=2))
|