Trajectory optimization using neural networks
Various embodiments of the technology described herein generally relate to systems and methods for trajectory optimization with machine learning techniques. More specifically, certain embodiments relate to using neural networks to quickly predict optimized robotic arm trajectories in a variety of scenarios. Systems and methods described herein use deep neural networks to quickly predict optimized robotic arm trajectories according to certain constraints. Optimization, in accordance with some embodiments of the present technology, may include optimizing trajectory geometry and dynamics while satisfying a number of constraints, including staying collision-free and minimizing the time it takes to complete the task.
1. A trajectory optimization method for a robotic system, the method comprising:
identifying a task for the robotic system, wherein the task comprises one or more requirements;
providing one or more details related to the task to a machine learning model, the machine learning model comprising a neural network trained to generate geometric trajectories for the robotic system that satisfy provided kinematic constraints and provided task constraints, wherein:
the machine learning model determines an optimal trajectory for the robotic system to follow when performing the task that satisfies the one or more requirements; and
the one or more details related to the task comprise a starting point and an ending point, but not a trajectory between the starting point and the ending point;
receiving an output from the machine learning model, wherein the output comprises the optimal trajectory, the optimal trajectory comprising a geometric path determined by the machine learning model; and
directing the robotic system to complete the task via the optimal trajectory.
2. The trajectory optimization method of claim 1 , wherein the one or more details related to the task are determined based on one or more images received from a computer vision system.
3. The trajectory optimization method of claim 1 , wherein the machine learning model is trained using one or more of reinforcement learning and imitation learning.
4. The trajectory optimization method of claim 1 , wherein the one or more details related to the task further comprise information about one or more obstacles between the starting point and the ending point.
5. The trajectory optimization method of claim 1 , wherein the one or more requirements include:
the robotic system does not collide with any objects;
a velocity of the robotic system remains below a specified maximum velocity; and
an acceleration of the robotic system remains below a specified maximum acceleration.
6. The trajectory optimization method of claim 1 , wherein the robotic system comprises one or more of: a robotic arm, a wheeled robot, and a self-driving car.
7. The trajectory optimization method of claim 1 , wherein:
the robotic system comprises a robotic arm; and
the task comprises picking up an object and moving the object to a new location.
8. An apparatus comprising:
one or more computer-readable storage media;
a processing system operatively coupled to the one or more computer-readable storage media; and
program instructions, stored on the one or more computer-readable storage media, wherein the program instructions, when read and executed by the processing system, direct the processing system to:
identify a task for a robotic system, wherein the task comprises one or more requirements;
provide one or more details related to the task to a machine learning model, the machine learning model comprising a neural network trained to generate geometric trajectories for the robotic system that satisfy provided kinematic constraints and provided task constraints, wherein:
the machine learning model determines an optimal trajectory for the robotic system to follow when performing the task that satisfies the one or more requirements; and
the one or more details related to the task comprise a starting point and an ending point, but not a trajectory between the starting point and the ending point;
receive an output from the machine learning model, wherein the output comprises the optimal trajectory, the optimal trajectory comprising a geometric path determined by the machine learning model; and
direct the robotic system to complete the task via the optimal trajectory.
9. The apparatus of claim 8 , wherein the one or more details related to the task are determined based on one or more images received from a computer vision system.
10. The apparatus of claim 8 , wherein the machine learning model is trained one or more of reinforcement learning and imitation learning.
11. The apparatus of claim 8 , wherein the one or more details related to the task further comprise information about one or more obstacles between the starting point and the ending point.
12. The apparatus of claim 8 , wherein the one or more requirements include:
the robotic system does not collide with any objects;
a velocity of the robotic system remains below a specified maximum velocity; and
an acceleration of the robotic system remains below a specified maximum acceleration.
13. The apparatus of claim 8 , wherein the robotic system comprises one or more of: a robotic arm, a wheeled robot, and a self-driving car.
14. The apparatus of claim 8 , wherein:
the robotic system comprises a robotic arm; and
the task comprises picking up an object and moving the object to a new location.
15. One or more non-transitory computer-readable storage media having program instructions stored thereon, wherein the program instructions, when read and executed by a processing system, direct the processing system to at least:
identify a task for a robotic system, wherein the task comprises one or more requirements;
provide one or more details related to the task to a machine learning model, the machine learning model comprising a neural network trained to generate geometric trajectories for the robotic system that satisfy provided kinematic constraints and provided task constraints, wherein:
the machine learning model determines an optimal trajectory for the robotic system to follow when performing the task that satisfies the one or more requirements; and
the one or more details related to the task comprise a starting point and an ending point, but not a trajectory between the starting point and the ending point;
receive an output from the machine learning model, wherein the output comprises the optimal trajectory, the optimal trajectory comprising a geometric path determined by the machine learning model; and
direct the robotic system to complete the task via the optimal trajectory.
16. The one or more non-transitory computer-readable storage media of claim 15 , wherein the one or more details related to the task are determined based on one or more images received from a computer vision system.
17. The one or more non-transitory computer-readable storage media of claim 15 , wherein the machine learning model is trained one or more of reinforcement learning and imitation learning.
18. The one or more non-transitory computer-readable storage media of claim 15 , wherein the one or more details related to the task further comprise information about one or more obstacles between the starting point and the ending point.
19. The one or more non-transitory computer-readable storage media of claim 15 , wherein the one or more requirements include:
the robotic system does not collide with any objects;
a velocity of the robotic system remains below a specified maximum velocity; and
an acceleration of the robotic system remains below a specified maximum acceleration.
20. The one or more non-transitory computer-readable storage media of claim 15 , wherein:
the robotic system comprises a robotic arm; and
the task comprises picking up an object and moving the object to a new location.