HongzeFu's picture
HF Space: code-only (no binary assets)
06c11b0
Raw
History Blame Contribute Delete
16.4 kB
import matplotlib.pyplot as plt
import matplotlib.patches as patches
from matplotlib.animation import FuncAnimation
import numpy as np
import torch
from ...logging_utils import logger
def grid_adjacency(R=3, C=3, diagonals=False, by_index=True):
"""
R, C: rows and columns
diagonals: False=4-connectivity; True=8-connectivity
by_index: True returns with index (0..R*C-1) as key; False returns with coordinate (r,c) as key
"""
# 4-connectivity directions
dirs = [(-1, 0), (1, 0), (0, -1), (0, 1)]
# 8-connectivity adds four diagonal directions
if diagonals:
dirs += [(-1, -1), (-1, 1), (1, -1), (1, 1)]
adj = {}
for r in range(R):
for c in range(C):
nbrs = []
for dr, dc in dirs:
nr, nc = r + dr, c + dc
if 0 <= nr < R and 0 <= nc < C:
nbrs.append((nr, nc))
key = r * C + c if by_index else (r, c)
if by_index:
adj[key] = [nr * C + nc for (nr, nc) in nbrs]
else:
adj[key] = nbrs
return adj
def dfs_path(adj, start, target, generator=None, blocked_nodes=None):
"""
Execute DFS to find path from start to target
Returns visit_order, path and edges used during search
Args:
adj: adjacency list
start: start node
target: target node
generator: torch.Generator for random neighbor selection
blocked_nodes: list of node indices to avoid (cannot pass through these nodes)
Returns:
visit_order: list of all nodes visited in order
path: final path from start to target
edges_used: edges traversed during search
"""
visited = set()
visit_order = [] # Track all nodes visited in order
path = []
edges_used = []
found = False
# Convert blocked_nodes to set for O(1) lookup
if blocked_nodes is None:
blocked_nodes = set()
else:
blocked_nodes = set(blocked_nodes)
def dfs_helper(node, current_path):
nonlocal found
if found:
return
visited.add(node)
visit_order.append(node) # Record visit order
current_path.append(node)
# Found target node
if node == target:
path.extend(current_path)
found = True
return
# Get neighbors and optionally shuffle them
neighbors = adj[node].copy()
if generator is not None:
# Use torch to randomly permute neighbors
neighbors_tensor = torch.tensor(neighbors)
perm = torch.randperm(len(neighbors), generator=generator)
neighbors = neighbors_tensor[perm].tolist()
# Continue searching
for neighbor in neighbors:
# Skip blocked nodes, already visited nodes
if neighbor not in visited and neighbor not in blocked_nodes and not found:
edges_used.append((node, neighbor))
dfs_helper(neighbor, current_path)
if found:
return
current_path.pop()
dfs_helper(start, [])
return visit_order, path, edges_used
def index_to_coord(idx, C):
"""Convert index to coordinate (row, col)"""
return idx // C, idx % C
def find_path_0_to_8(start, target, R=3, C=3, diagonals=False, generator=None, blocked_nodes=None):
"""Find path from start to target using DFS
Args:
start: start node
target: target node
R: number of rows
C: number of columns
diagonals: whether to use diagonal connections
generator: torch.Generator for random neighbor selection
blocked_nodes: list of node indices to avoid (cannot pass through these nodes)
Returns:
path: list of nodes in the final path from start to target (first return value)
visit_order: list of all nodes visited in order during DFS
edges_used: edges traversed during search
adj: adjacency list
"""
# Generate adjacency list
adj = grid_adjacency(R, C, diagonals=diagonals, by_index=True)
# Execute DFS to find path
visit_order, path, edges_used = dfs_path(adj, start, target, generator=generator, blocked_nodes=blocked_nodes)
if not path:
logger.debug(f"❌ Cannot find path from {start} to {target}!")
return None, None, None, None
# Only print the path
logger.debug(f"Path: {' β†’ '.join(map(str, path))}")
return path, visit_order, edges_used, adj
def run_path_generation(run_id, generator, segments, R=3, C=7, backtrack_enable=True):
"""
Run a single path generation with segments
Args:
run_id: Run identifier for printing
generator: torch.Generator for reproducible randomness
segments: List of segment tuples, each as (start, end, blocked_nodes)
e.g., [(7, 9, [8, 3, 10, 17]), (9, 11, [1, 8, 15]), ...]
R: Number of rows in grid
C: Number of columns in grid
backtrack_enable: Whether random backtracking passes are allowed
Returns:
combined_path: Complete path through all waypoints
combined_edges: All edges used in the path
adj: Adjacency list of the grid
start_node: Starting node (first segment's start)
end_node: Ending node (last segment's end)
Returns (None, None, None, None, None) if failed
"""
# Extract waypoints for display
waypoints = [segments[0][0]] # Start with first segment's start
for seg in segments:
waypoints.append(seg[1]) # Add each segment's end
logger.debug(f"\n{'=' * 60}")
logger.debug(f"Run {run_id}: Finding path through waypoints {' β†’ '.join(map(str, waypoints))}")
logger.debug(f"{'=' * 60}")
all_paths = []
all_edges = []
adj = None
for seg_idx, (seg_start, seg_target, seg_blocked) in enumerate(segments):
# Randomly decide whether to use backtracking for this segment
# Use torch generator for reproducible randomness
do_backtrack = backtrack_enable and (torch.rand(1, generator=generator).item() > 0.5)
# Print segment header
logger.debug(f"\nSegment {seg_idx + 1}: {seg_start} β†’ {seg_target}")
# Forward path: seg_start β†’ seg_target (always executed)
logger.debug(f" Path 1 (Forward {seg_start}β†’{seg_target}):")
path_forward, visit_order_fwd, edges_fwd, adj = find_path_0_to_8(
start=seg_start, target=seg_target, R=R, C=C, diagonals=False,
generator=generator, blocked_nodes=seg_blocked
)
if not path_forward:
logger.debug(f"\n❌ Failed to find forward path for segment {seg_idx + 1}, skipping this run")
return None, None, None, None, None
if do_backtrack:
# Backward path: seg_target β†’ seg_start (backtracking)
logger.debug(f" Path 2 (Backward {seg_target}β†’{seg_start}):")
path_backward, visit_order_bwd, edges_bwd, adj = find_path_0_to_8(
start=seg_target, target=seg_start, R=R, C=C, diagonals=False,
generator=generator, blocked_nodes=seg_blocked
)
if not path_backward:
logger.debug(f"\n❌ Failed to find backward path for segment {seg_idx + 1}, skipping this run")
return None, None, None, None, None
# Forward path again: seg_start β†’ seg_target
logger.debug(f" Path 3 (Forward {seg_start}β†’{seg_target}):")
path_forward2, visit_order_fwd2, edges_fwd2, adj = find_path_0_to_8(
start=seg_start, target=seg_target, R=R, C=C, diagonals=False,
generator=generator, blocked_nodes=seg_blocked
)
if not path_forward2:
logger.debug(f"\n❌ Failed to find second forward path for segment {seg_idx + 1}, skipping this run")
return None, None, None, None, None
# Combine: forward + backward + forward (removing duplicate nodes at connection points)
seg_combined = path_forward + path_backward[1:] + path_forward2[1:]
seg_edges = edges_fwd + edges_bwd + edges_fwd2
logger.debug(f" β†’ Segment generated 3 paths (with backtracking)")
else:
# No backtracking: just use forward path
seg_combined = path_forward
seg_edges = edges_fwd
logger.debug(f" β†’ Segment generated 1 path (no backtracking)")
all_paths.append(seg_combined)
all_edges.extend(seg_edges)
# Combine all paths (remove duplicate waypoint nodes between segments)
combined_path = all_paths[0]
for path_seg in all_paths[1:]:
combined_path = combined_path + path_seg[1:] # Skip first element (waypoint)
combined_edges = all_edges
logger.debug(f"\nβœ… Final combined path: {' β†’ '.join(map(str, combined_path))}")
# Return data separately (not as tuple)
return combined_path, combined_edges, adj, waypoints[0], waypoints[-1]
def visualize_single_path(path, edges_used, adj, start, target, R, C, blocked_nodes, run_id, segment_info):
"""
Visualize a single DFS path
Args:
path: list of nodes in the path
edges_used: list of edges used in the path
adj: adjacency dictionary
start: start node
target: target node
R: number of rows
C: number of columns
blocked_nodes: set of blocked node indices
run_id: run identifier for file naming
segment_info: string describing the segment selection (e.g., "segments[0:2]")
"""
if not path:
logger.debug("❌ No path to visualize!")
return
# Convert blocked_nodes to set for O(1) lookup
if blocked_nodes is None:
blocked_nodes = set()
else:
blocked_nodes = set(blocked_nodes)
fig, ax = plt.subplots(1, 1, figsize=(10, 10))
ax.set_xlim(-0.5, C - 0.5)
ax.set_ylim(-0.5, R - 0.5)
ax.set_aspect('equal')
ax.invert_yaxis()
ax.set_title(f'Run {run_id}: {segment_info}\nPath: {start} β†’ {target}',
fontsize=14, fontweight='bold', pad=20)
ax.set_xlabel('Column')
ax.set_ylabel('Row')
ax.grid(True, alpha=0.3)
# Draw all possible edges (adjacency relationships) - very light
for node, neighbors in adj.items():
r1, c1 = index_to_coord(node, C)
for neighbor in neighbors:
r2, c2 = index_to_coord(neighbor, C)
if node < neighbor:
ax.plot([c1, c2], [r1, r2], 'gray', alpha=0.15, linewidth=1, zorder=1)
# Draw path edges in blue
color = 'blue'
for i in range(len(path) - 1):
node = path[i]
neighbor = path[i + 1]
r1, c1 = index_to_coord(node, C)
r2, c2 = index_to_coord(neighbor, C)
ax.plot([c1, c2], [r1, r2], color=color, linewidth=3, alpha=0.7, zorder=3)
# Add arrows
dx, dy = c2 - c1, r2 - r1
ax.arrow(c1, r1, dx * 0.6, dy * 0.6,
head_width=0.12, head_length=0.08,
fc=color, ec=color, alpha=0.7, zorder=3)
# Draw all nodes
for idx in range(R * C):
r, c = index_to_coord(idx, C)
# Check if node is blocked
if idx in blocked_nodes:
node_color = 'dimgray'
else:
if idx == start:
node_color = 'lightgreen'
elif idx == target:
node_color = 'lightcoral'
elif idx in path:
node_color = 'lightyellow'
else:
node_color = 'lightgray'
circle = plt.Circle((c, r), 0.35, color=node_color, ec='black', linewidth=2, zorder=2)
ax.add_patch(circle)
ax.text(c, r, str(idx), ha='center', va='center',
fontsize=16, fontweight='bold',
color='white' if idx in blocked_nodes else 'black', zorder=10)
# Add X mark for blocked nodes
if idx in blocked_nodes:
ax.plot([c - 0.2, c + 0.2], [r - 0.2, r + 0.2], 'r', linewidth=3, zorder=11)
ax.plot([c - 0.2, c + 0.2], [r + 0.2, r - 0.2], 'r', linewidth=3, zorder=11)
# Add legend
from matplotlib.lines import Line2D
legend_elements = [
Line2D([0], [0], marker='o', color='w', markerfacecolor='lightgreen',
markersize=12, label='Start Node', markeredgecolor='black', markeredgewidth=2),
Line2D([0], [0], marker='o', color='w', markerfacecolor='lightcoral',
markersize=12, label='Target Node', markeredgecolor='black', markeredgewidth=2),
]
# Add blocked node indicator to legend if there are blocked nodes
if blocked_nodes:
legend_elements.append(
Line2D([0], [0], marker='o', color='w', markerfacecolor='dimgray',
markersize=12, label='Blocked Node', markeredgecolor='black', markeredgewidth=2)
)
ax.legend(handles=legend_elements, loc='upper left', fontsize=10, bbox_to_anchor=(1.02, 1))
plt.tight_layout()
# Save with run-specific filename
output_path = f'/Users/fuhongze/Desktop/robotic/verl/9grid/grid_path_run_{run_id}.png'
plt.savefig(output_path, dpi=150, bbox_inches='tight')
logger.debug(f"βœ… Visualization saved to: {output_path}")
plt.close(fig) # Close figure to free memory
if __name__ == "__main__":
print("=" * 60)
print("3x3 Grid: Multiple DFS Path Findings with Randomization")
print("=" * 60)
# Configuration
num_runs = 10 # You can change this number
# # Define segments as (start, end, blocked_nodes) tuples
segments = [
(7, 9, [8, 3, 10, 17]), # Segment 1: 7 β†’ 9
(9, 11, [1, 8, 15, 10, 5, 12, 19]), # Segment 2: 9 β†’ 11
(11, 13, [3, 10, 17, 12]) # Segment 3: 11 β†’ 13
]
segment_2 = [
(13, 11, [3, 10, 17, 12]), # Segment 3: 11 β†’ 13
(11, 9, [1, 8, 15, 10, 5, 12, 19]), # Segment 2: 9 β†’ 11
(9, 7, [8, 3, 10, 17]), # Segment 1: 7 β†’ 9
]
R, C = 3, 7
# Run multiple times and visualize each separately
for i in range(num_runs):
# Create generator with specific seed for this run
generator = torch.Generator()
generator.manual_seed(i)
# Randomly choose between segments and segment_2
segment_choice = torch.randint(0, 2, (1,), generator=generator).item()
chosen_segments = segments if segment_choice == 0 else segment_2
segment_name = "segments" if segment_choice == 0 else "segment_2"
# Randomly select a continuous slice from chosen segments
# Possible slices: [0:1], [1:2], [2:3], [0:2], [1:3], [0:3]
num_segments = len(chosen_segments)
start_idx = torch.randint(0, num_segments, (1,), generator=generator).item()
end_idx = torch.randint(start_idx + 1, num_segments + 1, (1,), generator=generator).item()
selected_segments = chosen_segments[start_idx:end_idx]
segment_info = f"{segment_name}[{start_idx}:{end_idx}]"
# Print selection
print(f"\n🎲 Selected: {segment_info}")
print(f" Segments to use:")
for idx, (seg_start, seg_target, seg_blocked) in enumerate(selected_segments, start=start_idx):
print(f" [{idx}] {seg_start} β†’ {seg_target}, blocked: {seg_blocked}")
# Collect all blocked nodes for visualization
all_blocked_nodes = set()
for _, _, blocked in selected_segments:
all_blocked_nodes.update(blocked)
# Run path generation
combined_path, combined_edges, adj, start_node, end_node = run_path_generation(
run_id=i+1,
generator=generator,
segments=selected_segments,
R=R,
C=C
)
# Visualize this run if generation was successful
if combined_path is not None:
print(f"\n{'=' * 60}")
print(f"Visualizing Run {i+1}")
print(f"{'=' * 60}")
visualize_single_path(
path=combined_path,
edges_used=combined_edges,
adj=adj,
start=start_node,
target=end_node,
R=R,
C=C,
blocked_nodes=all_blocked_nodes,
run_id=i+1,
segment_info=segment_info
)