Download astra/ClutterGrasp-v1/build_reference.py from Linslab/ManiLadder_Models: direct link, hf CLI and curl.
- Browser
- Download file 1.83 kB
-
https://huggingface.co/Linslab/ManiLadder_Models/resolve/main/astra/ClutterGrasp-v1/build_reference.py
- Command line
-
hf download hf://Linslab/ManiLadder_Models/astra/ClutterGrasp-v1/build_reference.py
-
curl -L -o build_reference.py https://huggingface.co/Linslab/ManiLadder_Models/resolve/main/astra/ClutterGrasp-v1/build_reference.py
1.83 kB
| """Offline reference compilation from the 20 authorized demos only. No fitting.""" | |
| import importlib.util,json,sys | |
| from pathlib import Path | |
| import numpy as np | |
| sys.path.insert(0,str(Path(__file__).resolve().parent)) | |
| from controller import image_features,CAMERAS | |
| s=importlib.util.spec_from_file_location('worker_api','/worker_api.py');m=importlib.util.module_from_spec(s);s.loader.exec_module(m);api=m.API() | |
| manifest=json.load(open('/demo_manifest.json')) | |
| assert len(manifest['episodes'])==20 | |
| rows={k:[] for k in ['features','qpos','qvel','actions','demo','frame']};audit=[] | |
| for di,e in enumerate(manifest['episodes']): | |
| n=e['length'];fields=['actions','obs/agent/qpos','obs/agent/qvel'];d=api.read_demo(e['demo_id'],fields=fields) | |
| indices=list(range(21,n,2));audit.append({'demo_id':e['demo_id'],'computational_rgb_frames':indices,'label_intervals':[[t,min(t+2,n)] for t in indices]}) | |
| for start in range(21,n,64): | |
| stop=min(n,start+64) | |
| imgs=api.read_demo(e['demo_id'],start,stop,fields=['obs/sensor_data/'+c+'/rgb' for c in CAMERAS]) | |
| for t in range(start,stop,2): | |
| rows['features'].append(image_features({c:imgs['obs/sensor_data/'+c+'/rgb'][t-start] for c in CAMERAS})) | |
| rows['qpos'].append(d['obs/agent/qpos'][t]);rows['qvel'].append(d['obs/agent/qvel'][t]) | |
| rows['actions'].append(d['actions'][t:min(t+2,n)].mean(0));rows['demo'].append(di);rows['frame'].append(t) | |
| print(e['demo_id'],len(indices),flush=True) | |
| b={k:np.asarray(v,dtype=np.int32 if k in ['demo','frame'] else np.float32) for k,v in rows.items()} | |
| b['qscale']=np.maximum(np.std(b['qpos'],axis=0),.20);b['vscale']=np.maximum(np.std(b['qvel'],axis=0),.10) | |
| np.savez_compressed(Path(__file__).with_name('reference_bank.npz'),**b) | |
| json.dump(audit,open(Path(__file__).with_name('reference_index.json'),'w'),indent=2) | |
| print('bank',len(b['actions']),flush=True) | |