Deep Reinforcement Learning Approach for Small Spacecraft Manipulator Collision Avoidance