Reliable Deployment of Machine Learning for Long-Term Autonomy
IROS 2020 - Workshop
Achieving long-term autonomy by mobile robots means the ability to operate autonomously under no/or minimal supervision for days, weeks, months or even years. During these long periods, the environment where the robot operates can experience unpredictable gradual or/and radical changes. This fact adds an extra dimension to the fundamental problems in robotics such as perception, planning, navigation, SLAM and manipulation; and makes them more challenging.
One of the keys to achieving long-term autonomy is having reliable sub-components in the robotic operating system, including the machine learning-based onse. In this context, reliability means that the components can identify and recover from failures and prevent or reduce the likelihood of failures in general, which otherwise can terminate the mission of the robot or/and might cause severe danger.
This workshop focuses on the problem of long-term autonomy for mobile robots and the challenge of building a reliabile machine learning components in the robotic system that can handle bad sensory data, shifts to abnormal operational conditions, misclassification and detections.
We invite several renowned experts in the field who will highlight the main challenges these robots face and talk about their own experiences and the lessons they learnt during long-term deployments of their robots.
Reasoning about environmental appearance and structural change.
Lifelong learning and adaptation.
Failure detection and recovery.
Long-term mission planning and exploration
Spatial representation for long-term mapping and localisation.
State estimation in dynamic environments.
Context-dependent decision making
Verification of long-term autonomous systems
Reliability, Dependability and Explainability of Machine Learning for robotics.
Connor Basich, Justin Svegliato and Shlomo Zilberstein, Improving Competence for Long-Term Autonomy.
Matthias Humt, Jongseok Lee and Rudolph Triebel, Bayesian Optimization Meets Laplace Approximation for Robotic Introspection.
Deeksha Dixit and Pratap Tokekar, Evaluation of Cross-View Matching to Improve Ground Vehicle Localization with Aerial Perception.
Feras Dayoub, Australian Centre for Robotic Vision, Queensland University of Technology, AU
Tomas Krajnik, Czech Technical University in Prague
Niko Suenderhauf, Australian Centre for Robotic Vision, Queensland University of Technology, AU
Ayoung Kim, Korea Advanced Institute of Science and Technology