tostido's picture
Add blueprints archive: ARACHNE-001, MARIONETTE-001, AIRFOIL-CORDAGE-SYSTEM, PERSPECTIVE
26fa66a
"""
Cable Geometry & Intersection Detection
========================================
Handles the REAL physics of tether geometry:
- Each TAB has a defined operational SECTOR
- Cables cannot cross without consequences
- Intersection detection using line-line distance
- Tangling physics when cables touch
- Forced release or system drag when tangled
The 4 TABs are in cross formation:
UP (z+)
|
|
LEFT ---+--- RIGHT (y axis)
(y-) |
|
DOWN (z-)
^ Forward (x axis - direction of flight)
Each TAB operates in a WEDGE-SHAPED SECTOR:
- UP: angles 45° to 135° from vertical (upper hemisphere)
- DOWN: angles -45° to -135° (lower hemisphere)
- LEFT: angles 135° to 225° (left hemisphere)
- RIGHT: angles -45° to 45° (right hemisphere)
Cable intersection occurs when line segments cross in 3D space.
"""
import numpy as np
from dataclasses import dataclass
from typing import Dict, List, Tuple, Optional, Set
from enum import Enum
import itertools
class TangleState(Enum):
"""Cable tangling states"""
CLEAR = "clear" # No issues
PROXIMITY = "proximity" # Getting close - warning
CROSSED = "crossed" # Cables have crossed
TANGLED = "tangled" # Wrapped around each other
LOCKED = "locked" # Cannot be separated without release
@dataclass
class CableGeometry:
"""
Represents a single cable from mother drone to TAB.
The cable is modeled as a line segment from anchor to TAB position,
with some sag based on tension.
"""
tab_id: str
anchor_point: np.ndarray # Point on mother drone
tab_position: np.ndarray # Current TAB position
cable_length: float # Maximum cable length
current_tension: float = 0.0 # Current tension (N)
@property
def direction(self) -> np.ndarray:
"""Unit vector from anchor to TAB"""
vec = self.tab_position - self.anchor_point
norm = np.linalg.norm(vec)
if norm < 1e-6:
return np.array([1.0, 0.0, 0.0])
return vec / norm
@property
def extension(self) -> float:
"""Current cable extension (may be less than max)"""
return np.linalg.norm(self.tab_position - self.anchor_point)
@property
def slack(self) -> float:
"""How much slack in the cable (length - extension)"""
return max(0, self.cable_length - self.extension)
def get_midpoint(self) -> np.ndarray:
"""Get cable midpoint (for intersection checks)"""
return (self.anchor_point + self.tab_position) / 2
def sample_points(self, n: int = 10) -> np.ndarray:
"""
Sample points along the cable for intersection checking.
Includes simple catenary sag model when cable has slack.
"""
t = np.linspace(0, 1, n)
points = np.zeros((n, 3))
for i, ti in enumerate(t):
# Linear interpolation
p = self.anchor_point + ti * (self.tab_position - self.anchor_point)
# Add sag based on slack and position
# Maximum sag at midpoint
if self.slack > 0.1:
sag_factor = 4 * ti * (1 - ti) # Parabolic
sag_amount = self.slack * 0.3 * sag_factor
p[2] -= sag_amount # Sag downward
points[i] = p
return points
@dataclass
class OperationalSector:
"""
Defines the allowed operational wedge for a TAB.
The sector is defined by angle ranges in the YZ plane
(perpendicular to flight direction).
Angle 0 = +Y (right)
Angle 90 = +Z (up)
Angle 180/-180 = -Y (left)
Angle -90 = -Z (down)
"""
tab_id: str
angle_min: float # Radians
angle_max: float # Radians
radial_min: float = 5.0 # Minimum distance from center (m)
radial_max: float = 35.0 # Maximum (cable length)
def contains_angle(self, angle: float) -> bool:
"""Check if angle is within sector"""
# Normalize angle to [-pi, pi]
angle = np.arctan2(np.sin(angle), np.cos(angle))
# Handle wrap-around
if self.angle_min <= self.angle_max:
return self.angle_min <= angle <= self.angle_max
else:
# Sector crosses -pi/pi boundary
return angle >= self.angle_min or angle <= self.angle_max
def contains_point(self, point_yz: np.ndarray) -> bool:
"""Check if a YZ point is within the sector"""
y, z = point_yz[0], point_yz[1]
r = np.sqrt(y**2 + z**2)
angle = np.arctan2(z, y)
if r < self.radial_min or r > self.radial_max:
return False
return self.contains_angle(angle)
def clamp_to_sector(self, point_yz: np.ndarray) -> np.ndarray:
"""Project a point to the nearest valid sector location"""
y, z = point_yz[0], point_yz[1]
r = np.sqrt(y**2 + z**2)
angle = np.arctan2(z, y)
# Clamp radius
r = np.clip(r, self.radial_min, self.radial_max)
# Clamp angle to sector
if not self.contains_angle(angle):
# Find nearest sector boundary
d_min = self._angle_distance(angle, self.angle_min)
d_max = self._angle_distance(angle, self.angle_max)
angle = self.angle_min if d_min < d_max else self.angle_max
return np.array([r * np.cos(angle), r * np.sin(angle)])
def _angle_distance(self, a1: float, a2: float) -> float:
"""Compute minimum angular distance"""
diff = a2 - a1
return abs(np.arctan2(np.sin(diff), np.cos(diff)))
# Define the operational sectors for each TAB position
# Sectors have a small gap between them to prevent intersection
OPERATIONAL_SECTORS = {
"UP": OperationalSector(
tab_id="UP",
angle_min=np.radians(50), # 50° from +Y
angle_max=np.radians(130), # 130° from +Y
),
"DOWN": OperationalSector(
tab_id="DOWN",
angle_min=np.radians(-130), # -130° from +Y
angle_max=np.radians(-50), # -50° from +Y
),
"LEFT": OperationalSector(
tab_id="LEFT",
angle_min=np.radians(140), # 140° from +Y (into left side)
angle_max=np.radians(-140), # Wraps around -180
),
"RIGHT": OperationalSector(
tab_id="RIGHT",
angle_min=np.radians(-40), # -40° from +Y
angle_max=np.radians(40), # 40° from +Y
),
}
class CableIntersectionDetector:
"""
Detects and manages cable intersection geometry.
Key responsibilities:
1. Track all cable geometries
2. Detect when cables cross
3. Compute intersection severity
4. Determine consequences (warning, drag, forced release)
"""
# Minimum distance between cables before intersection warning
PROXIMITY_THRESHOLD = 2.0 # meters
INTERSECTION_THRESHOLD = 0.5 # meters - cables touching
def __init__(self, cable_length: float = 30.0):
self.cable_length = cable_length
self.cables: Dict[str, CableGeometry] = {}
self.tangle_states: Dict[Tuple[str, str], TangleState] = {}
self.sectors = OPERATIONAL_SECTORS.copy()
# Track intersection history for tangling
self.intersection_history: Dict[Tuple[str, str], List[float]] = {}
def update_cable(self,
tab_id: str,
mother_pos: np.ndarray,
tab_pos: np.ndarray,
tension: float = 0.0):
"""Update cable geometry for a TAB"""
# Anchor points on mother drone (offset from center)
anchor_offsets = {
"UP": np.array([0, 0, 2]),
"DOWN": np.array([0, 0, -2]),
"LEFT": np.array([0, -3, 0]),
"RIGHT": np.array([0, 3, 0]),
}
anchor = mother_pos + anchor_offsets.get(tab_id, np.zeros(3))
self.cables[tab_id] = CableGeometry(
tab_id=tab_id,
anchor_point=anchor,
tab_position=tab_pos,
cable_length=self.cable_length,
current_tension=tension
)
def check_all_intersections(self) -> Dict[Tuple[str, str], float]:
"""
Check for intersections between all cable pairs.
Returns dict of (tab1, tab2) -> minimum distance
"""
distances = {}
cable_ids = list(self.cables.keys())
for i, id1 in enumerate(cable_ids):
for id2 in cable_ids[i+1:]:
dist = self._compute_cable_distance(id1, id2)
distances[(id1, id2)] = dist
# Update tangle state
self._update_tangle_state(id1, id2, dist)
return distances
def _compute_cable_distance(self, id1: str, id2: str) -> float:
"""
Compute minimum distance between two cables.
Uses line segment to line segment distance.
"""
cable1 = self.cables[id1]
cable2 = self.cables[id2]
# Get endpoints
p1, p2 = cable1.anchor_point, cable1.tab_position
p3, p4 = cable2.anchor_point, cable2.tab_position
return self._segment_segment_distance(p1, p2, p3, p4)
def _segment_segment_distance(self,
p1: np.ndarray, p2: np.ndarray,
p3: np.ndarray, p4: np.ndarray) -> float:
"""
Compute minimum distance between two line segments.
Uses the algorithm from:
"Distance between Lines and Segments with their Closest Point of Approach"
"""
d1 = p2 - p1 # Direction of segment 1
d2 = p4 - p3 # Direction of segment 2
r = p1 - p3
a = np.dot(d1, d1)
e = np.dot(d2, d2)
f = np.dot(d2, r)
# Check if both segments are points
if a < 1e-8 and e < 1e-8:
return np.linalg.norm(p1 - p3)
if a < 1e-8:
# Segment 1 is a point
s = 0.0
t = np.clip(f / e, 0, 1)
else:
c = np.dot(d1, r)
if e < 1e-8:
# Segment 2 is a point
t = 0.0
s = np.clip(-c / a, 0, 1)
else:
b = np.dot(d1, d2)
denom = a * e - b * b
if abs(denom) > 1e-8:
s = np.clip((b * f - c * e) / denom, 0, 1)
else:
s = 0.0
t = (b * s + f) / e
if t < 0:
t = 0
s = np.clip(-c / a, 0, 1)
elif t > 1:
t = 1
s = np.clip((b - c) / a, 0, 1)
closest1 = p1 + s * d1
closest2 = p3 + t * d2
return np.linalg.norm(closest1 - closest2)
def _update_tangle_state(self, id1: str, id2: str, distance: float):
"""Update the tangle state between two cables"""
key = tuple(sorted([id1, id2]))
current = self.tangle_states.get(key, TangleState.CLEAR)
if distance > self.PROXIMITY_THRESHOLD:
new_state = TangleState.CLEAR
elif distance > self.INTERSECTION_THRESHOLD:
new_state = TangleState.PROXIMITY
else:
# Cables are intersecting
if current in (TangleState.CLEAR, TangleState.PROXIMITY):
new_state = TangleState.CROSSED
elif current == TangleState.CROSSED:
# Track how long they've been crossed
history = self.intersection_history.get(key, [])
history.append(distance)
self.intersection_history[key] = history
if len(history) > 10: # Crossed for multiple updates
new_state = TangleState.TANGLED
else:
new_state = TangleState.CROSSED
elif current == TangleState.TANGLED:
new_state = TangleState.LOCKED
else:
new_state = current
# Clear history if cables separate
if new_state == TangleState.CLEAR:
if key in self.intersection_history:
del self.intersection_history[key]
self.tangle_states[key] = new_state
def get_tangle_state(self, id1: str, id2: str) -> TangleState:
"""Get current tangle state between two cables"""
key = tuple(sorted([id1, id2]))
return self.tangle_states.get(key, TangleState.CLEAR)
def get_tangled_pairs(self) -> List[Tuple[str, str]]:
"""Get list of cable pairs that are tangled or locked"""
return [
pair for pair, state in self.tangle_states.items()
if state in (TangleState.TANGLED, TangleState.LOCKED)
]
def get_crossed_pairs(self) -> List[Tuple[str, str]]:
"""Get list of cable pairs that are currently crossed"""
return [
pair for pair, state in self.tangle_states.items()
if state in (TangleState.CROSSED, TangleState.TANGLED, TangleState.LOCKED)
]
def is_in_sector(self, tab_id: str, position_yz: np.ndarray) -> bool:
"""Check if TAB position is within its allowed sector"""
if tab_id not in self.sectors:
return True
return self.sectors[tab_id].contains_point(position_yz)
def clamp_to_sector(self, tab_id: str, position_yz: np.ndarray) -> np.ndarray:
"""Clamp position to valid sector"""
if tab_id not in self.sectors:
return position_yz
return self.sectors[tab_id].clamp_to_sector(position_yz)
def compute_drag_penalty(self) -> float:
"""
Compute additional drag from cable tangling.
Tangled cables create turbulence and drag.
"""
penalty = 0.0
for pair, state in self.tangle_states.items():
if state == TangleState.CROSSED:
penalty += 0.1 # Minor drag increase
elif state == TangleState.TANGLED:
penalty += 0.5 # Significant drag
elif state == TangleState.LOCKED:
penalty += 1.0 # Major drag, system unstable
return penalty
def get_forced_releases(self) -> List[str]:
"""
Get TABs that must be released due to locked cables.
When cables are LOCKED, one must be released to prevent
catastrophic drag on the mother drone.
"""
locked_pairs = [
pair for pair, state in self.tangle_states.items()
if state == TangleState.LOCKED
]
if not locked_pairs:
return []
# Release the TABs involved in locks
# Strategy: release the one with lower tension (less load)
releases = set()
for id1, id2 in locked_pairs:
c1 = self.cables.get(id1)
c2 = self.cables.get(id2)
if c1 and c2:
if c1.current_tension <= c2.current_tension:
releases.add(id1)
else:
releases.add(id2)
elif c1:
releases.add(id1)
elif c2:
releases.add(id2)
return list(releases)
class SectorConstrainedActionSpace:
"""
Constrains TAB actions to valid sectors.
This ensures the agent can NEVER request an action that would
cause cable intersection - the impossible maneuvers are removed
from the action space entirely.
"""
def __init__(self, cable_length: float = 30.0):
self.cable_length = cable_length
self.sectors = OPERATIONAL_SECTORS.copy()
# Build action masks per TAB
self.action_masks = self._compute_action_masks()
def _compute_action_masks(self) -> Dict[str, np.ndarray]:
"""
Precompute valid action regions for each TAB.
Actions are (elevator, rudder) which translate to
(pitch, yaw) which determine YZ position.
"""
masks = {}
# Discretize action space for masking
# Elevator affects Z, Rudder affects Y
for tab_id, sector in self.sectors.items():
# Create mask over discretized YZ positions
resolution = 20
y_range = np.linspace(-self.cable_length, self.cable_length, resolution)
z_range = np.linspace(-self.cable_length, self.cable_length, resolution)
mask = np.zeros((resolution, resolution), dtype=bool)
for i, y in enumerate(y_range):
for j, z in enumerate(z_range):
if sector.contains_point(np.array([y, z])):
mask[i, j] = True
masks[tab_id] = mask
return masks
def constrain_action(self,
tab_id: str,
action_yz: np.ndarray,
current_pos_yz: np.ndarray) -> np.ndarray:
"""
Constrain an action to stay within the TAB's valid sector.
Args:
tab_id: Which TAB
action_yz: Requested action in YZ plane (relative motion)
current_pos_yz: Current YZ position
Returns:
Constrained action that won't leave sector
"""
if tab_id not in self.sectors:
return action_yz
sector = self.sectors[tab_id]
# Compute target position
target_yz = current_pos_yz + action_yz
# If target is valid, allow it
if sector.contains_point(target_yz):
return action_yz
# Otherwise, clamp to sector boundary
clamped = sector.clamp_to_sector(target_yz)
return clamped - current_pos_yz
def get_valid_action_range(self,
tab_id: str,
current_pos_yz: np.ndarray) -> Tuple[np.ndarray, np.ndarray]:
"""
Get the valid action range for a TAB at current position.
Returns (min_action, max_action) bounds.
"""
if tab_id not in self.sectors:
return (np.array([-10, -10]), np.array([10, 10]))
sector = self.sectors[tab_id]
# Sample boundary and find limits
angles = np.linspace(sector.angle_min, sector.angle_max, 50)
boundary_points = []
for angle in angles:
for r in [sector.radial_min, sector.radial_max]:
y = r * np.cos(angle)
z = r * np.sin(angle)
boundary_points.append(np.array([y, z]))
boundary = np.array(boundary_points)
# Compute action bounds (target - current)
deltas = boundary - current_pos_yz
return deltas.min(axis=0), deltas.max(axis=0)
def get_sector_boundaries_for_viz() -> Dict[str, List[np.ndarray]]:
"""
Get sector boundary lines for visualization.
Returns dict of tab_id -> list of line points
"""
viz_data = {}
for tab_id, sector in OPERATIONAL_SECTORS.items():
points = []
# Inner arc
angles = np.linspace(sector.angle_min, sector.angle_max, 30)
for angle in angles:
y = sector.radial_min * np.cos(angle)
z = sector.radial_min * np.sin(angle)
points.append(np.array([0, y, z])) # At x=0 (cross-section)
# Outer arc (reverse for closed polygon)
for angle in reversed(angles):
y = sector.radial_max * np.cos(angle)
z = sector.radial_max * np.sin(angle)
points.append(np.array([0, y, z]))
# Close the shape
if points:
points.append(points[0])
viz_data[tab_id] = points
return viz_data