Constrained Dynamic Movement Primitives for Collision-Safe Robots
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robotic systems using Dynamic Movement Primitives (DMPs) face challenges in incorporating constraints during task execution, such as collision avoidance and joint limits, leading to potential self-collision or environmental collisions.
Innovation Solution
The method involves transforming DMPs into Constrained DMPs (CDMPs) by defining a perturbation function that adapts the forcing function to satisfy operational constraints. This is achieved through a non-linear optimization problem that optimizes the weights of the basis functions, allowing the robot to operate within the defined constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional DMP-based techniques are used for trajectory generation, then the robot can learn complex trajectories from demonstrations with generalization capabilities, but the robot may collide with itself or the environment or extend past its joint limits when constraints are present
Solution Approach 1:
The patent applies preliminary action by pre-defining constraint functions that represent collision avoidance and joint limits before trajectory execution. These constraint functions are incorporated into the DMP formulation in advance, allowing the robot to automatically satisfy constraints when executing learned trajectories without requiring real-time correction or relearning.
Solution Approach 2:
The patent changes parameters by modifying the DMP forcing function parameters to include constraint satisfaction terms. By adjusting the parameters of the basis functions and the forcing function, the system generates trajectories that inherently respect collision avoidance and joint limit constraints while maintaining the learned skill's essential characteristics.
2Reliability
If trajectory correction is performed when constraints are violated, then collision avoidance may be achieved, but the correction is computationally expensive, impractical, and may be impossible when the generated trajectory severely violates the constraints
Solution Approach 1:
The patent applies preliminary action by pre-defining constraint functions that represent collision avoidance and joint limits before trajectory execution. These constraint functions are incorporated into the DMP formulation in advance, allowing the robot to automatically satisfy constraints when executing learned trajectories without requiring real-time correction or relearning.
Solution Approach 2:
The patent extracts the constraint satisfaction problem from the trajectory correction process and incorporates it directly into the DMP forcing function. By taking out the need for separate correction algorithms and embedding constraint handling within the DMP framework itself, the system achieves computationally efficient constraint satisfaction.
3Adaptability or versatility
If the forcing function is relearned for different environments, then the robot can adapt to new constraints, but relearning is impractical when environment changes occur during task execution
Solution Approach 1:
The patent applies universality by creating a DMP forcing function formulation that can handle multiple different environments and constraint configurations using the same learned skill. The constraint functions are designed to be environment-specific parameters rather than requiring relearning of the entire forcing function, allowing the same learned skill to be universally applied across different environments by simply adjusting constraint parameters.
Data Source
AI summary
A controller for controlling an operation of a robot to execute a task is provided. The controller comprises a memory configured to store a set of dynamic movement primitives (DMPs) associated with the task. The set of DMPs comprise a set of at least two dynamical systems: a function representing point attractor dynamics and a forcing function corresponding to a learned demonstration of the task. The controller comprises a processor configured to transform the set of DMPs to a set of constrained DMPs (CDMPs) by determining a perturbation function associated with the forcing function. The perturbation function is associated with a set of operational constraints. The processor is further configured to solve, a non-linear optimization problem for the set of CDMPs based on the set of operational constraints and generate, a control input for controlling the robot for executing the task, based on the solution.