Uncertainty-Aware Learning from Demonstration using Mixture Density\n Networks with Sampling-Free Variance Modeling

2017/09/03 by Sungjoon Choi, Choi, Sungjoon, Kyungjae Lee +5 · 4 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · Engineering · #Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Fault Detection and Control Systems #Gaussian Processes and Bayesian Inference #Gene Regulatory Network Analysis #Machine Learning (cs.LG) #Robotics (cs.RO)

paper · pdf · doi:10.48550/arxiv.1709.02249

openalex publication_date 2017/09/03 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

In this paper, we propose an uncertainty-aware learning from demonstration\nmethod by presenting a novel uncertainty estimation method utilizing a mixture\ndensity network appropriate for modeling complex and noisy human behaviors. The\nproposed uncertainty acquisition can be done with a single forward path without\nMonte Carlo sampling and is suitable for real-time robotics applications. The\nproperties of the proposed uncertainty measure are analyzed through three\ndifferent synthetic examples, absence of data, heavy measurement noise, and\ncomposition of functions scenarios. We show that each case can be distinguished\nusing the proposed uncertainty measure and presented an uncertainty-aware\nlearn- ing from demonstration method of an autonomous driving using this\nproperty. The proposed uncertainty-aware learning from demonstration method\noutperforms other compared methods in terms of safety using a complex\nreal-world driving dataset.\n

Cited by

• Where to Measure: Epistemic Uncertainty-Based Sensor Placement with ConvCNPs
• Uncertainty-Aware Autonomous Vehicles: Predicting the Road Ahead
• Adaptive Individual Uncertainty under Out-Of-Distribution Shift with Expert-Routed Conformal Prediction
• Learning Social Navigation from Positive and Negative Demonstrations and Rule-Based Specifications

Related

• Uncertainty Aware Learning from Demonstrations in Multiple Contexts\n using Bayesian Neural Networks
• Uncertainty Estimation based on Geometric Separation
• Bridging Active Exploration and Uncertainty-Aware Deployment Using Probabilistic Ensemble Neural Network Dynamics
• DADEE: Well-calibrated uncertainty quantification in neural networks for barriers-based robot safety
• Learning an Uncertainty-Aware Object Detector for Autonomous Driving