|
Algorithm 1 Real-Time Parallel Trajectory Optimization |
-
1:
Input: Previous trajectory , control inputs (warm-start)
-
2:
while within time budget and convergence not met do
-
3:
Parallel: For each knot along the trajectory:
-
4:
Compute gradients for smoothness, obstacle cost, and dynamics
-
5:
Compute constraint gradients (e.g., collision or control limits)
-
6:
Aggregate gradients
-
7:
Update and (e.g., via projected gradient descent)
-
8:
Project updated trajectory to enforce hard constraints
-
9:
end while
-
10:
Output: Refined trajectory , controls
|