We are used to see “armies” of robots manufacturing cars or handling packages at sorting facilities. These are very structured environments, with restricted access, where robots can be programmed in a deterministic way. In order to move to unstructured environments, and have them sharing their workspace with humans, we need to account for uncertainties and develop stochastic, adaptive control schemes.
Active inference is a stepping stone we are currently using to address this challenge. Active Inference is prominent in the neuroscientific literature as a general theory of the human brain (Friston, 2010). We can loosely describe it here as the tendency of our brain to both update its own beliefs about the world around us, and act upon it in a way that minimizes “surprise”. In particular, such surprise is defined through the so-called Free Energy and depends on the difference between an internal generative model of the word held by the brain, and sensory information.
Several recent approaches in robotics have taken inspiration from it (Buckley et al., 2017) for tasks such as state-estimation (Baioumy et al., 2021), adaptive control (Pezzato et al., 2020), predictive control (Baioumy et al., 2020), human-robot interaction (Ohata & Tani, 2020), reinforcement learning (Tschantz et al., 2020) and many more.
Together with TUD colleagues Corrado Pezzato and Carlos Hernandez Corbato, and Mohamed Bayoumi and Nick Hawes from Oxford Robotics Institute, we recently proposed a novel fault tolerant controller for robotic manipulators based on Active Inference (Pezzato et al., 2020; Baioumy et al., 2021). We are currently working on extending it to learn online the accuracy of its own sensors, which can be used to adapt to natural degradations or complete failures (Baioumy et al., 2021).
Joint work with (mostly): Corrado Pezzato, Carlos Hernandez Corbato, Mohamed Bayoumi and Nick Hawes.
Publications
A Novel Adaptive Controller for Robot Manipulators based on Active Inference
Pezzato, Corrado,
Ferrari, Riccardo M.G.,
and Hernández Corbato, Carlos
IEEE Robotics and Automation Letters
2020
Unbiased Active Inference for Classical Control
Baioumy, Mohamed,
Pezzato, Corrado,
Ferrari, Riccardo M.G.,
and Hawes, Nick
In International Conference on Intelligent Robots and Systems
2022
Active inference is a mathematical framework that originated in computational neuroscience. Recently, it has been demonstrated as a promising approach for constructing goal-driven behavior in robotics. Specifically, the active inference controller (AIC) has been successful on several continuous control and state-estimation tasks. Despite its relative success, some established design choices lead to a number of practical limitations for robot control. These include having a biased estimate of the state, and only an implicit model of control actions. In this paper, we highlight these limitations and propose an extended version of the unbiased active inference controller (u-AIC). The u-AIC maintains all the compelling benefits of the AIC and removes its limitations. Simulation results on a 2-DOF arm and experiments on a real 7-DOF manipulator show the improved performance of the u-AIC with respect to the standard AIC. The code can be found at https://github.com/cpezzato/unbiased_aic.
Beta Residuals: Improving Fault-Tolerant Control for Sensory Faults via Bayesian Inference and Precision Learning
Baioumy, Mohamed,
Hartemink, William,
Ferrari, Riccardo M.G.,
and Hawes, Nick
In IFAC Symposium on Fault Detection, Supervision and Safety for Technical Processes
2022
Model-based fault-tolerant control (FTC) often consists of two distinct steps: fault detection & isolation (FDI), and fault accommodation. In this work we investigate posing fault- tolerant control as a single Bayesian inference problem. Previous work showed that precision learning allows for stochastic FTC without an explicit fault detection step. While this leads to implicit fault recovery, information on sensor faults is not provided, which may be essential for triggering other impact-mitigation actions. In this paper, we introduce a precision-learning based Bayesian FTC approach and a novel beta residual for fault detection. Simulation results are presented, supporting the use of beta residual against competing approaches.
Towards Stochastic Fault-tolerant Control using Precision Learning and Active Inference
Baioumy, Mohamed,
Pezzato, Corrado,
Corbato, Carlos Hernández,
Hawes, Nick,
and Ferrari, Riccardo M.G.
In International Workshop on Active Inference
2021
Fault-tolerant Control of Robotic Systems with Sensory Faults using Unbiased Active Inference
Baioumy, Mohamed,
Pezzato, Corrado,
Ferrari, Riccardo M.G.,
Corbato, Carlos Hernández,
and Hawes, Nick
In European Control Conference
2021
Active inference for fault tolerant control of robot manipulators with sensory faults
Pezzato, Corrado,
Baioumy, Mohamed,
Corbato, Carlos Hernández,
Hawes, Nick,
Wisse, Martijn,
and Ferrari, Riccardo M.G.
In International Workshop on Active Inference
2020
A Novel Adaptive Controller for Robot Manipulators based on Active Inference
Pezzato, Corrado,
Ferrari, Riccardo M.G.,
and Hernández Corbato, Carlos
In International Conference on Robotics and Automation
2020
Additional References
The free-energy principle: a unified brain theory?
Friston, K. J.
Nature Reviews Neuroscience
2010
The free energy principle for action and perception: A mathematical review
Buckley, Christopher L,
Kim, Chang Sub,
McGregor, Simon,
and Seth, Anil K
Journal of Mathematical Psychology
2017
Active Inference for Integrated State-Estimation, Control, and Learning
Baioumy, Mohamed,
Duckworth, Paul,
Lacerda, Bruno,
and Hawes, Nick
In Proc of IEEE Int. conference on robotics and automation (ICRA)
2021
Variational Inference for Predictive and Reactive Controllers
Baioumy, Mohamed,
Mattamala, Matias,
and Hawes, Nick
In Workshop on New advances in Brain-inspired Perception, Interaction and Learning
2020
Investigation of the Sense of Agency in Social Cognition, Based on Frameworks of Predictive Coding and Active Inference: A Simulation Study on Multimodal Imitative Interaction
Ohata, Wataru,
and Tani, Jun
Frontiers in Neurorobotics
2020
Reinforcement learning through active inference
Tschantz, Alexander,
Millidge, Beren,
Seth, Anil K,
and Buckley, Christopher L
2020