Robot Motion Planning via Unified Contact Sequence and Pose Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multi-contact motion planning methods in robotics are inefficient due to the separation of contact sequence and pose estimation steps, leading to suboptimal trajectory planning and limited adaptation of contact poses, which results in computationally intensive and less effective motion planning.
Innovation Solution
A method that acquires an initial sequence of postures with contact points and kinematic poses, modifies constraint topologies, generates constraint equations, performs relaxation, and applies a trajectory generation algorithm to optimize effector trajectories, enabling more efficient and adaptive motion planning by considering the overall sequence of postures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If the problem of multi-contact motion planning is divided into separate subtasks (contact sequence search and pose estimation), then the computational complexity of each individual step is reduced, but the interdependencies between steps are lost leading to suboptimal overall solutions
Solution Approach 1:
The patent merges the previously separate contact sequence search and pose estimation steps into a unified optimization framework. The cost function simultaneously optimizes both the contact sequence and contact poses, allowing interdependencies to be preserved while maintaining computational tractability through efficient optimization algorithms.
Solution Approach 2:
The patent introduces dynamic adaptation of contact poses based on the optimized contact sequence. The pose estimation is no longer static but dynamically adjusted according to the optimal contact sequence found, enabling the system to adapt poses for a given contact while considering the overall sequence optimality.
2Adaptability or versatility
If contact poses are adapted using traditional Inverse Kinematics techniques between optimization steps, then some pose adjustment is achieved, but the adaptation is limited and does not consider overall sequence optimality
Solution Approach 1:
The patent combines pose adaptation and sequence optimization into a single unified process. Instead of adapting poses separately after sequence determination, the cost function simultaneously optimizes both aspects, achieving higher adaptability while improving computational efficiency by eliminating multiple separate optimization passes.
3Reliability
If ample computation resources are allocated to find optimal multi-contact motion planning solutions, then solution quality improves, but computational time and resource requirements increase significantly
Solution Approach 1:
The patent performs preliminary optimization of the cost function that incorporates both contact sequence and pose estimation. By pre-defining the optimization framework and cost function structure, the system achieves high solution quality without requiring excessive computational resources during execution, as the heavy lifting is done through efficient gradient-based optimization.
Solution Approach 2:
The patent changes the optimization parameters by using a unified cost function that directly optimizes both contact sequence and poses. This parameter transformation allows the use of efficient continuous optimization methods rather than discrete search methods, significantly reducing computational time while maintaining solution quality.
Data Source
AI summary
A method for controlling at least one effector trajectory of an effector of a robot for solving a predefined task is proposed. A sequence of postures are acquired to modify at least one of a contact constraint topology and an object constraint topology. A set of constraint equations are generated based on at least one of the modified contact constraint topology and the modified object constraint topology. On the generated set of constraint equations, a constraint relaxation is performed to generate a task description including a set of relaxed constraint equations. The at least one effector trajectory is generated by applying a trajectory generation algorithm on the task description. An inverse kinematics algorithm is performed to generate a control signal from the at least one effector trajectory. At least one effector is controlled to execute the at least one effector trajectory based on the generated control signal.