Online Robot Introspection and Decision Making

This project explores how to perform multimodal robot skill acquisition, in particular for contact tasks. We explore novel ways of modeling, integrating, and classifying low-level robot sensory-motor information or high-level abstracted representations. Our approaches help bootstrap a robot’s ability to perform introspection into the kinds of behaviors it executes as it runs. This in turns aids in the ability to recognize unmodeled and external disturbances that will likely lead the task to fail, and instead, learn efficient recovery strategies to continue with the task at hand.

In this project, we are exploring how to use heuristics, Bayesian nonparametrics, Markov switching processes, variational inference, and deep reinforcement learning on single and dual-arm robots to perform robust contact tasks that might be suitable for service and industry.

Related Work:

Decision Making in Recovery Scenarios

• Error Identification and Recovery in Robotic Snap Assembly
• Recovering from External Disturbances in Manipulation through Re-Enactment and Adaptation
• Recovering from External Disturbances in Online Manipulation through State-Dependent Revertive Recovery Policies

Hidden Markov Models and Markov Switching Process Techniques

• A Latent State-based Multimodal Execution Monitor with Anomaly Detection and Classification for Robot Introspection
• Analysis of multimodal Bayesian nonparametric autoregressive hidden Markov models for process monitoring in robotic contact tasks
• Fast, Robust, and Versatile Event Detection through HMM Belief State Gradient Measures
• Robot Introspection with Bayesian Nonparametric Vector Autoregressive Hidden Markov Models

Supervised Machine Learning

• Robot Introspection via Wrench-based Action Grammars (offline) and (online)
• Robot Contact Task State Estimation via Position-based Action Grammars

Heuristics

• Contextualized Early Failure Characterization of Cantilever Snap Assemblies
• Early Failure Characterization of Cantilever Snap Assemblies
• Cantilever Snap Assemblies Failure Detection using SVMs and the RCBHT
• Towards Snap Sensing
• A Gradient Calibration for the RCBHT Cantilever Snap Verification System
• Probabilistic State Verification for Snap Assemblies using the Relative-Change-Based Hierarchical Taxonomy
• A Relative-Change-Based Hierarchical Taxonomy for Cantilever-Snap Assembly Verification