Cascading Variational Autoencoder for Robot Collision Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Robots currently lack effective methods to detect collisions during tasks, which can lead to damage to themselves or environmental objects, as existing systems fail to accurately identify collisions with unseen or misrecognized objects or external events like falling objects.

Innovation Solution

A cascading variational autoencoder (VAE) network is used to determine collisions by processing robot states and actions, comparing predicted and actual latent spaces to identify discrepancies that indicate a collision, allowing for corrective actions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional collision detection methods are used, then the system is simple, but collision detection accuracy is poor and cannot detect unseen or misrecognized objects

Engineering Contradiction:

Improvecollision detection accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces a cascading VAE network as an intermediary between robot state sensing and collision detection. The network includes multiple VAEs that process robot states through encoder-decoder architectures with latent space representations, enabling indirect detection of collisions through learned patterns rather than direct sensing, thus improving detection accuracy while maintaining computational feasibility

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces traditional mechanical or rule-based collision detection systems with a data-driven deep learning approach. The VAE network learns collision patterns from training data and uses probabilistic inference to detect collisions, substituting deterministic mechanical sensing with stochastic computational modeling that can detect subtle collision signatures

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If a robot performs tasks with end effectors, then productivity is improved, but the risk of collision damage to components and environmental objects increases

Engineering Contradiction:

Improvetask execution capabilityVSAvoidcollision damage risk

Core Design Contradiction:

ProductivityVSObject-affected harmful factors

Solution Approach 1:

The patent implements continuous feedback monitoring through the cascading VAE network that processes robot states in real-time during task execution. The network compares predicted versus actual robot states and generates collision probability estimates, providing ongoing feedback that enables dynamic adjustment of robot behavior to prevent collision damage while maintaining productivity

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent trains the VAE network in advance with extensive training data that includes both normal operation and collision scenarios. This preliminary training enables the network to learn collision patterns and establish baseline behaviors before actual task execution, allowing the robot to detect and respond to collisions proactively rather than reactively

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11285604B2Robot collision detection using cascading variational autoencoder

Publication Date: 2022.03.29 GDM HOLDING LLC

AI summary

Cascading variational autoencoder (“VAE”) models can be used to detect robot collisions while a robot is performing a task. For a current state of the robot, various implementations include a VAE used to generate a latent space of the current state, and a predictor network used to generate a predicted latent space for the current state. A collision can be determined based on a difference between the latent space for the current state and the predicted latent space for the current state.