Force-Torque Robotic Assembly Across Platforms Without Pose Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional robotic control techniques struggle to adapt to unstructured environments, such as construction sites, due to the need for precise calibration and pose tracking, which are often unreliable and require robot-specific training, limiting their applicability and adaptability.
Innovation Solution
A machine learning model trained via reinforcement learning uses force and torque measurements from sensors to control robotic assembly tasks, without requiring pose tracking, and is agnostic to the specific robot type, allowing it to generalize across different robotic platforms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional learning-based techniques use motion capture or tracking systems to capture pose, then control accuracy is improved, but system complexity and calibration difficulty increase
Solution Approach 1:
The patent extracts and removes the pose capture and tracking systems from the robotic assembly system. Instead of using motion capture cameras or external tracking devices, the invention uses only force sensors mounted on the robot end-effector to sense contact forces and torques during assembly operations. This eliminates the complexity of calibrating and deploying motion capture systems while maintaining control accuracy through force-based feedback.
Solution Approach 2:
The patent replaces the optical/mechanical pose tracking system with a force-based sensing system. Instead of using cameras to visually track robot position and orientation, the invention uses force sensors to detect mechanical interactions between the robot and assembly components. This substitution simplifies the system by eliminating complex optical tracking infrastructure while providing direct tactile feedback about assembly contact states.
2Ease of manufacture
If vision-based systems are used to infer pose indirectly, then system deployment is simplified, but measurement reliability deteriorates in contact-rich phases with occlusion and poor lighting
Solution Approach 1:
The patent replaces vision-based pose inference with force-based sensing. Instead of relying on cameras to visually determine robot position and orientation (which fail in occluded or poorly lit conditions), the invention uses force sensors to directly measure mechanical contact forces and torques. This mechanical sensing approach is immune to lighting conditions and occlusions, providing reliable feedback throughout the entire assembly process including contact-rich phases.
3Manufacturing precision
If conventional techniques are used to control robots, then control precision is maintained, but adaptability to unstructured environments deteriorates
Solution Approach 1:
The patent implements force feedback control where sensors on the robot end-effector continuously measure contact forces and torques during assembly operations. This real-time force feedback is fed back to the controller, which adjusts robot movements to maintain precise control while adapting to variations in component positioning, tolerances, and environmental conditions. The feedback loop enables the system to handle unstructured environments while maintaining manufacturing precision through active compensation.
Solution Approach 2:
The patent transitions from static, pre-programmed control to dynamic, adaptive control based on real-time force sensing. The control system continuously adjusts robot trajectory and applied forces based on feedback from force sensors, enabling the robot to adapt to unexpected variations in the assembly environment while maintaining precise control. This dynamic approach allows the same control system to handle both structured and unstructured environments effectively.
4Measurement precision
If robot-specific training is used, then control accuracy for that robot is improved, but generalizability to other robotic platforms deteriorates
Solution Approach 1:
The patent develops a universal force-based control framework that can be applied to different robotic platforms without robot-specific training. By using force sensors mounted on the end-effector and implementing control algorithms based on contact force feedback rather than platform-specific kinematics, the system achieves both high control accuracy and broad generalizability. The force sensing approach is platform-agnostic, allowing the same control software to work with different robot types while maintaining precision through adaptive force control.
Data Source
AI summary
Techniques are disclosed for training and applying machine learning models to control robotic assembly. In some embodiments, force and torque measurements are input into a machine learning model that includes a memory layer that introduces recurrency. The machine learning model is trained, via reinforcement learning in a robot-agnostic environment, to generate actions for achieving an assembly task given the force and torque measurements. During training, experiences are collected as transitions within episodes, the transitions are grouped into sequences, and the last two sequences of each episode have a variable overlap. The collected transitions are stored in a prioritized sequence replay buffer, from which a learner samples sequences to learn from based on transition and sequence priorities. Once trained, the machine learning model can be deployed to control various types of robots to perform the assembly task based on force and torque measurements acquired by sensors of those robots.