Datasets:
Download code/dynamic_dataset/geometry.py from Heiheihaha17/MESH-Benchmark: direct link, hf CLI and curl.
- Browser
- Download file 4.96 kB
-
https://huggingface.co/datasets/Heiheihaha17/MESH-Benchmark/resolve/main/code/dynamic_dataset/geometry.py
- Command line
-
hf download hf://datasets/Heiheihaha17/MESH-Benchmark/code/dynamic_dataset/geometry.py
-
curl -L -o geometry.py https://huggingface.co/datasets/Heiheihaha17/MESH-Benchmark/resolve/main/code/dynamic_dataset/geometry.py
4.96 kB
| """Continuous minimum-jerk motion, swept footprints and query ambiguity checks.""" | |
| from __future__ import annotations | |
| import copy,math | |
| import numpy as np | |
| from .common import SR,active,geometry,grid,position | |
| from dynamic_demo.render import visible | |
| def points_clear(points,room,margin=.25): | |
| points=np.asarray(points);w,h,d=room['room_size_m'] | |
| if not ((np.abs(points[...,0])<w/2-margin)&(np.abs(points[...,2])<d/2-margin)&(points[...,1]>.2)&(points[...,1]<h-.2)).all():return False | |
| for box in room['surfaces'][6:]: | |
| if np.any(np.all(np.abs(points[...,[0,2]]-np.asarray(box['location'])[[0,2]])<np.asarray(box['size'])[[0,2]]/2+margin,axis=-1)):return False | |
| return True | |
| def initial_positions(entities,room,rng,force=False): | |
| """Keep parent positions when valid; otherwise sample legal body footprints.""" | |
| w,_,d=room['room_size_m'];placed=[];reused=True | |
| for e in entities: | |
| point=np.asarray(e['initial_xyz_m']);others=placed | |
| valid=not force and points_clear(point,room,.35) and all(np.linalg.norm((point-q)[[0,2]])>=.9 for q in others) | |
| if not valid: | |
| reused=False | |
| for _ in range(3000): | |
| point=np.array([rng.uniform(-w/2+.5,w/2-.5),point[1],rng.uniform(-d/2+.5,d/2-.5)]) | |
| if points_clear(point,room,.35) and all(np.linalg.norm((point-q)[[0,2]])>=.9 for q in others):break | |
| else:raise ValueError('Cannot place entities in this room') | |
| e['initial_xyz_m']=point.tolist();placed.append(point) | |
| return reused | |
| def range_queries(scene): | |
| queries=[] | |
| for human in scene['humans']: | |
| anchor=next(s/SR for s in grid(scene) if active(human,s)) | |
| for receiver in scene['robots']: | |
| az=geometry(human,receiver,anchor)['doa_deg'][0];lo=int(math.floor(az/30))*30;hi=lo+30 | |
| rounded=copy.deepcopy(receiver);rounded['initial_xyz_m']=[round(float(x),1) for x in position(receiver,anchor)] | |
| rounded['displacement_m']=[0.,0.,0.];rounded['yaw_deg']=round(receiver['yaw_deg']) | |
| azq=geometry(human,rounded,anchor)['doa_deg'][0] | |
| if not (lo+1<az<hi-1 and lo+1<azq<hi-1):return None | |
| selected=[h['entity_id'] for h in scene['humans'] if active(h,round(anchor*SR)) and lo<=geometry(h,receiver,anchor)['doa_deg'][0]<hi] | |
| if selected!=[human['entity_id']]:return None | |
| queries.append(dict(source_id=human['entity_id'],receiver_id=receiver['entity_id'], | |
| selection_time_s=anchor,cue_time_range_s=[anchor,round(anchor+.2,9)],initial_sector_deg=[lo,hi])) | |
| return queries | |
| def audit(scene,room,full_visibility=True): | |
| cues=range_queries(scene) | |
| if cues is None:return None | |
| ts=np.arange(0,scene['duration_samples']/SR,.025);entities=scene['humans']+scene['robots'] | |
| pts=[position(e,ts) for e in entities];minimum=float('inf') | |
| for e,p in zip(entities,pts): | |
| if not points_clear(p,room):return None | |
| length=float(np.linalg.norm(e['displacement_m']));peak=1.875*length/e['motion_duration_s'] | |
| if length and (length<=2. or peak>.8000001):return None | |
| if e.get('motion_start_s',0)+e['motion_duration_s']>scene['duration_samples']/SR:return None | |
| for i,p in enumerate(pts): | |
| for q in pts[i+1:]: | |
| minimum=min(minimum,float(np.linalg.norm((p-q)[:,[0,2]],axis=1).min())) | |
| if minimum<.6:return None | |
| # Reception points move; use their current positions for every visibility check. | |
| if full_visibility: | |
| for human in scene['humans']: | |
| for t in ts[::4]: | |
| if not any(visible(position(human,t),position(r,t),room) for r in scene['robots']):return None | |
| observability=[] | |
| for h in scene['humans']: | |
| active_ts=[s/SR for s in grid(scene) if active(h,s)] | |
| for r in scene['robots']: | |
| az=np.unwrap(np.radians([geometry(h,r,t)['doa_deg'][0] for t in active_ts]));observability.append(dict(source_id=h['entity_id'],receiver_id=r['entity_id'],active_azimuth_span_deg=float(np.degrees(np.ptp(az))))) | |
| if scene['branch']=='easy_human': | |
| first=scene['humans'][0]['entity_id'];r0=scene['robots'][0]['entity_id'] | |
| if next(x['active_azimuth_span_deg'] for x in observability if x['source_id']==first and x['receiver_id']==r0)<30:return None | |
| elif scene['branch']=='easy_robot': | |
| r0=scene['moving_entity_ids'][0] | |
| if max(x['active_azimuth_span_deg'] for x in observability if x['receiver_id']==r0)<30:return None | |
| else: | |
| if any(max(x['active_azimuth_span_deg'] for x in observability if x['source_id']==h['entity_id'])<15 for h in scene['humans']):return None | |
| return dict(minimum_horizontal_clearance_m=minimum,observability=observability,range_queries=cues, | |
| displacement_m={e['entity_id']:float(np.linalg.norm(e['displacement_m'])) for e in entities}, | |
| maximum_speed_mps=max(1.875*np.linalg.norm(e['displacement_m'])/e['motion_duration_s'] for e in entities)) | |