Trajectory Control of a Computer Arm

TL;DR

This paper presents a systematic approach to programming a computer-controlled arm, dividing it into planning and execution phases. It introduces a method for generating smooth, collision-free trajectories and a servoing technique based on Lagrangian mechanics to accurately follow these trajectories, enabling complex manipulation tasks.

What Problem Does This Paper Address?

The paper addresses the challenge of programming robotic arms to perform complex manipulation tasks. Traditional approaches often required specialized programming for each task. This paper aims to provide a more general and systematic method by separating the task into two distinct phases: planning (deciding what to do and generating a trajectory) and execution (moving the arm along the trajectory and performing hand actions). A key difficulty lies in generating smooth, safe, and efficient trajectories that account for the arm's dynamics, physical constraints, and potential collisions with the environment. Furthermore, accurately controlling the arm to follow these complex trajectories in real-time is a significant engineering challenge.

Methods

The paper outlines a two-phase approach:

  • Planning:
    • Trajectory Generation: The core of planning involves generating a time-dependent sequence of joint positions, velocities, and accelerations. This is achieved using polynomial functions.
  • Task Decomposition: Tasks are broken down into segments, such as moving to an object, grasping, moving the object, and returning to a rest position.
  • Collision Avoidance: Trajectories are checked for collisions with physical stops and other objects. A heuristic approach is used to modify trajectories to avoid obstacles, prioritizing vertical ascent and descent.
  • Hand Actions: Simple hand actions (open, close, grasp) are specified as procedures to be executed at specific points in the trajectory.
  • Execution (Servoing):
    • Lagrangian Mechanics: The servo program uses Lagrangian mechanics to model the arm's dynamics, accounting for joint coupling and varying inertial loading.
  • Sampled Data System: The servo operates as a sampled data system with a 60 Hz sample rate.
  • Torque Calculation: Joint accelerations are calculated based on position and velocity errors. These accelerations are then used with the arm's dynamic model to compute the required joint torques.
  • Motor Control: Torques are converted into motor commands (voltage pulse width) by considering gear ratios, efficiency, back EMF, and motor resistance.

Datasets

Not discussed in the paper. The paper focuses on the methodology and implementation rather than empirical evaluation on specific datasets.

Key Insights and Ideas

  • Separation of Planning and Execution: This fundamental division simplifies the programming process, allowing for modular development and easier modification of tasks.
  • Polynomial Trajectory Generation: Using piecewise polynomials provides a smooth and controllable way to define arm motion, ensuring continuity of position, velocity, and acceleration.
  • Dynamic Model for Servoing: Incorporating Lagrangian mechanics for servoing is crucial for achieving accurate control, especially in a complex, multi-jointed arm where joint interactions are significant.
  • Heuristic Collision Avoidance: While not a perfect solution, the heuristic approach for collision avoidance demonstrates a practical way to handle unexpected obstacles during motion.
  • Systematic Approach: The paper advocates for a systematic, rather than ad-hoc, method for arm programming, reducing the need for task-specific code.

Key Results

The paper reports successful servoing of the arm along trajectories resulting in a change of position of approximately 40 inches in less than 1 second, with a final accuracy of 0.02% or 1/30 inch. This demonstrates the effectiveness of their trajectory generation and servoing methods.

Limitations and Future Work

  • Real-time Planning Limitations: The paper acknowledges that planning a trajectory is severely limited in time (1-2 seconds), which can be a constraint for highly complex tasks.
  • Heuristic Nature of Collision Avoidance: While functional, the heuristic collision avoidance might not be optimal or robust for all scenarios, especially with complex object shapes or tight spaces.
  • Hardware Dependency: The servoing method relies on accurate knowledge of the arm's physical parameters and the performance of the computer and its interfaces.
  • Future Work: The paper implies that further refinement of planning algorithms, more sophisticated collision avoidance strategies, and potentially integration with higher-level reasoning systems would be valuable future directions. The authors also suggest that the computer's role in servoing has made modification of approaches simple, allowing for experimentation with various techniques.

Summary updated

BibTeX
@inproceedings{1971trajectory,
  title = {{Trajectory Control of a Computer Arm}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1971},
}
Figure