Modeling Motor Control in Continuous Time Active Inference: A Survey

IEEE Transactions on Cognitive and Developmental Systems, Journal Year: 2023, Volume and Issue: 16(2), P. 485 - 500

Published: Dec. 4, 2023

The way the brain selects and controls actions is still widely debated. Mainstream approaches based on Optimal Control focus stimulus-response mappings that optimize cost functions. Ideomotor theory cybernetics propose a different perspective: they suggest are selected controlled by activating action effects continuously matching internal predictions with sensations. Active Inference offers modern formulation of these ideas, in terms inferential mechanisms prediction-error-based control, which can be linked to neural living organisms. This article provides technical illustration models continuous time brief survey solve four kinds control problems; namely, goal-directed reaching movements, active sensing, resolution multisensory conflict during movement integration decision-making motor control. Crucially, Inference, all facets emerge from same optimization process - minimization Free Energy do not require designing separate Therefore, unitary perspective various aspects inform both study biological design artificial robotic systems.

Language: Английский

---

Symmetry and complexity in object-centric deep active inference models

Interface Focus, Journal Year: 2023, Volume and Issue: 13(3)

Published: April 14, 2023

Humans perceive and interact with hundreds of objects every day. In doing so, they need to employ mental models these often exploit symmetries in the object's shape appearance order learn generalizable transferable skills. Active inference is a first principles approach understanding modeling sentient agents. It states that agents entertain generative model their environment, act by minimizing an upper bound on surprisal, i.e. Free Energy. The Energy decomposes into accuracy complexity term, meaning favor least complex model, can accurately explain sensory observations. this paper, we investigate how inherent particular also emerge as latent state space learnt under deep active inference. particular, focus object-centric representations, which are trained from pixels predict novel object views agent moves its viewpoint. First, relation between symmetry exploitation space. Second, do principal component analysis demonstrate encodes axis Finally, more symmetrical representations be exploited for better generalization context manipulation.

Language: Английский

---

Deep kinematic inference affords efficient and scalable control of bodily movements

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: May 5, 2023

ABSTRACT Performing goal-directed movements requires mapping goals from extrinsic (workspace-relative) to intrinsic (body-relative) coordinates and then motor signals. Mainstream approaches based on Optimal Control realize the mappings by minimizing cost functions, which is computationally demanding. Instead, Active Inference uses generative models produce sensory predictions, allows a cheaper inversion However, devising control complex kinematic chains like human body challenging. We introduce novel architecture that affords simple but effective via inference easily scales up drive chains. Rich can be specified in both using attractive or repulsive forces. The proposed model reproduces sophisticated bodily paves way for efficient biologically plausible of actuated systems.

Language: Английский

---

An Overview of the Free Energy Principle and Related Research

Neural Computation, Journal Year: 2024, Volume and Issue: 36(5), P. 963 - 1021

Published: March 8, 2024

Abstract The free energy principle and its corollary, the active inference framework, serve as theoretical foundations in domain of neuroscience, explaining genesis intelligent behavior. This states that processes perception, learning, decision making—within an agent—are all driven by objective “minimizing energy,” evincing following behaviors: learning employing a generative model environment to interpret observations, thereby achieving selecting actions maintain stable preferred state minimize uncertainty about environment, making. fundamental can be used explain how brain perceptual information, learns selects actions. Two pivotal tenets are agent employs for perception planning interaction with world (and other agents) enhances performance augments perception. With evolution control theory deep tools, agents based on FEP have been instantiated various ways across different domains, guiding design multitude models decision-making algorithms. letter first introduces basic concepts FEP, followed historical development connections theories intelligence, then delves into specific application making, encompassing both low-dimensional simple situations high-dimensional complex situations. It compares model-based reinforcement show provides better function. We illustrate this using numerical studies Dreamer3 adding expected information gain standard In complementary fashion, existing algorithms also help implement FEP-based agents. Finally, we discuss capabilities need possess environments aid acquiring these capabilities.

Language: Английский

---

Oscillating latent dynamics in robot systems during walking and reaching

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: May 19, 2024

Abstract Sensorimotor control of complex, dynamic systems such as humanoids or quadrupedal robots is notoriously difficult. While artificial traditionally employ hierarchical optimisation approaches black-box policies, recent results in neuroscience suggest that complex behaviours locomotion and reaching are correlated with limit cycles the primate motor cortex. A result suggests that, when applied to a learned latent space, oscillating patterns activation can be used physical robot. reminiscent observed cortex, these dynamics unsurprising given cyclic nature robot’s behaviour (walking). In this preliminary investigation, we consider how similar approach extends less obviously (reaching). This has been explored prior work using computational simulations. But simulations necessarily make simplifying assumptions do not correspond reality, so trivially transfer real robot platforms. Our primary contribution demonstrate infer states learnt representation oscillatory during tasks. We further show encodes interpretable movements workspace. Compared locomotion, observe for fully cyclic, they begin end at same position space. However, trace out shape cycle, and, by construction, driven underlying mechanics.

Language: Английский

---

Dynamics of an information theoretic analog of two masses on a spring

Chaos Solitons & Fractals, Journal Year: 2024, Volume and Issue: 188, P. 115535 - 115535

Published: Sept. 17, 2024

Language: Английский

---

Incremental Learning of Goal-Directed Actions in a Dynamic Environment by a Robot Using Active Inference

Entropy, Journal Year: 2023, Volume and Issue: 25(11), P. 1506 - 1506

Published: Oct. 31, 2023

This study investigated how a physical robot can adapt goal-directed actions in dynamically changing environments, real-time, using an active inference-based approach with incremental learning from human tutoring examples. Using our model, while good generalization be achieved appropriate parameters, when faced sudden, large changes the environment, may have to intervene correct of order reach goal, as caregiver might guide hands child performing unfamiliar task. In for learn tutor, we propose new scheme accomplish these proprioceptive–exteroceptive experiences combined mental rehearsal past experiences. Our experimental results demonstrate that only few examples, model was able significantly improve its performance on tasks without catastrophic forgetting previously learned tasks.

Language: Английский

---

Recursive neural programs: A differentiable framework for learning compositional part-whole hierarchies and image grammars

PNAS Nexus, Journal Year: 2023, Volume and Issue: 2(11)

Published: Oct. 14, 2023

Human vision, thought, and planning involve parsing representing objects scenes using structured representations based on part-whole hierarchies. Computer vision machine learning researchers have recently sought to emulate this capability neural networks, but a generative model formulation has been lacking. Generative models that leverage compositionality, recursion, hierarchies are thought underlie human concept the ability construct represent flexible mental concepts. We introduce Recursive Neural Programs (RNPs), addresses hierarchy problem by modeling images as hierarchical trees of probabilistic sensory-motor programs. These programs recursively reuse learned primitives an image within different spatial reference frames, enabling composition from parts implementing grammar for images. show RNPs can learn variety datasets, allowing rich compositionality intuitive parts-based explanations objects. Our also suggests cognitive framework understanding how brains potentially concepts in terms defined their relations with each other.

Language: Английский

---

Human–Artificial Intelligence Systems: How Human Survival First Principles Influence Machine Learning World Models

Systems, Journal Year: 2022, Volume and Issue: 10(6), P. 260 - 260

Published: Dec. 17, 2022

World models is a construct that used to represent internal of the world. It an important for human-artificial intelligence systems, because both natural and artificial agents can have world models. The term, agents, encompasses individual people human organizations. Many organizations apply include machine learning. In this paper, it explained how survival first principles interactions between energy entropy influence organization’s models, hence their implementations First, related This done in terms construct’s origins psychology theory-building during 1930s through its applications systems science 1970s recent computational neuroscience. Second, organizational Third, practical example provided lead opposing Fourth, constrain Overall, paper highlights on organizations’ doing so, profound challenges are revealed systems.

Language: Английский

---

The Emperor Is Naked: Replies to commentaries on the target article

Behavioral and Brain Sciences, Journal Year: 2022, Volume and Issue: 45

Published: Jan. 1, 2022

Abstract The 35 commentaries cover a wide range of topics and take many different stances on the issues explored by target article. We have organised our response to around three central questions: Are Friston blankets just Pearl blankets? What ontological metaphysical commitments are implied use kind explanatory work capable of? conclude reply with short critical reflection indiscriminate both Markov free energy principle.

Language: Английский

---