Intuitive Physics Research Articles

Tangled Physics: Knots Strain Intuitive Physical Reasoning.

Whereas decades of research have cataloged striking errors in physical reasoning, a resurgence of interest in intuitive physics has revealed humans' remarkable ability to successfully predict the unfolding of physical scenes. A leading interpretation intended to resolve these opposing results is that physical reasoning recruits a general-purpose mechanism that reliably models physical scenarios (explaining recent successes), but overly contrived tasks or impoverished and ecologically invalid stimuli can produce poor performance (accounting for earlier failures). But might there be tasks that persistently strain physical understanding, even in naturalistic contexts? Here, we explore this question by introducing a new intuitive physics task: evaluating the strength of knots and tangles. Knots are ubiquitous across cultures and time-periods, and evaluating them correctly often spells the difference between safety and peril. Despite this, 5 experiments show that observers fail to discern even very large differences in strength between knots. In a series of two-alternative forced-choice tasks, observers viewed a variety of simple "bends" (knots joining two pieces of thread) and decided which would require more force to undo. Though the strength of these knots is well-documented, observers' judgments completely failed to reflect these distinctions, across naturalistic photographs (E1), idealized renderings (E2), dynamic videos (E3), and even when accompanied by schematic diagrams of the knots' structures (E4). Moreover, these failures persisted despite accurate identification of the topological differences between the knots (E5); in other words, even when observers correctly perceived the underlying structure of the knot, they failed to correctly judge its strength. These results expose a blindspot in physical reasoning, placing new constraints on general-purpose theories of scene understanding.

Perceiving animacy through schematic intuitive physics: Shared conceptual structure of animacy between vision and language

Perceiving animacy through schematic intuitive physics: Shared conceptual structure of animacy between vision and language

Unconscious intuitive physics: Prioritized breakthrough into visual awareness for physically unstable block towers

Unconscious intuitive physics: Prioritized breakthrough into visual awareness for physically unstable block towers

Temporal segmentation and "look ahead" simulation: Physical events structure visual perception of intuitive physics.

How we perceive the physical world is not only organized in terms of objects, but also structured in time as sequences of events. This is especially evident in intuitive physics, with temporally bounded dynamics such as falling, occlusion, and bouncing demarcating the continuous flow of sensory inputs. While the spatial structure and attentional consequences of physical objects have been well-studied, much less is known about the temporal structure and attentional consequences of physical events in visual perception. Previous work has recognized physical events as units in the mind, and used presegmented object interactions to explore physical representations. However, these studies did not address whether and how perception imposes the kind of temporal structure that carves these physical events to begin with, and the attentional consequences of such segmentation during intuitive physics. Here, we use performance-based tasks to address this gap. In Experiment 1, we find that perception not only spontaneously separates visual input in time into physical events, but also, this segmentation occurs in a nonlinear manner within a few hundred milliseconds at the moment of the event boundary. In Experiment 2, we find that event representations, once formed, use coarse "look ahead" simulations to selectively prioritize those objects that are predictively part of the unfolding dynamics. This rich temporal and predictive structure of physical event representations, formed during vision, should inform models of intuitive physics. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Characterizing Chinese undergraduate students' empathizing-systemizing profiles: a person-centered approach.

While the empathizing-systemizing (E-S) theory provides a valuable framework for explaining gender differences in STEM majors, previous studies suffer from methodological issues (i.e., the arbitrary cut-off criteria and WEIRD sampling) as well as discrepancies in the behavioral correlates of E-S types. To address the gaps, this study utilized a 3-step latent profile analysis to identify naturally occurring E-S profiles in a Chinese sample and explored the predictors and distal outcomes of the identified profiles. The study recruited 785 (aged 18-25 years, 60% female) Chinese undergraduates. Results revealed five E-S profiles: Disengaged, Empathizers, Navigating systemizers, Technological systemizers, and Self-declared allrounders. Controlling for socioeconomic status, being male predicted a higher likelihood of membership into the Technological systemizers. Besides, membership to the Navigating systemizers and Technological systemizers was associated with better intuitive physics performance. However, no significant variation was observed for social sensitivity performance across E-S profiles. Overall, our results partially conformed to previous findings, highlighting the importance of cultural adaptation and methodological considerations when classifying students' cognitive types.

ASIMO: Agent-centric scene representation in multi-object manipulation

Vision-based reinforcement learning (RL) is a generalizable way to control an agent because it is agnostic of specific hardware configurations. As visual observations are highly entangled, attempts for vision-based RL rely on scene representation that discerns individual entities and establishes intuitive physics to constitute the world model. However, most existing works on scene representation learning cannot successfully be deployed to train an RL agent, as they are often highly unstable and fail to sustain for a long enough temporal horizon. We propose ASIMO, a fully unsupervised scene decomposition to perform interaction-rich tasks with a vision-based RL agent. ASIMO decomposes agent-object interaction videos of episodic-length into the agent, objects, and background, predicting their long-term interactions. Further, we explicitly model possible occlusion in the image observations and stably track individual objects. Then, we can correctly deduce the updated positions of individual entities in response to the agent action, only from partial visual observation. Based on the stable entity-wise decomposition and temporal prediction, we formulate a hierarchical framework to train the RL agent that focuses on the context around the object of interest. We demonstrate that our formulation for scene representation can be universally deployed to train different configurations of agents and accomplish several tasks that involve pushing, arranging, and placing multiple rigid objects.

Exploring Alternative Approaches to Language Modeling for Learning from Data and Knowledge

Despite their extensive application in language understanding tasks, large language models (LLMs) still encounter challenges including hallucinations - occasional fabrication of information - and alignment issues - lack of associations with human-curated world models (e.g., intuitive physics or common-sense knowledge). Moreover, the black-box nature of LLMs presents significant obstacles in training them effectively to achieve desired behaviors. In particular, modifying the concept embedding spaces of LLMs can be highly intractable. This process involves analyzing the implicit impact of such adjustments on the myriad parameters within LLMs and the resulting inductive biases. We propose a novel architecture that wraps powerful function approximation architectures within an outer, interpretable read-out layer. This read-out layer can be scrutinized to explicitly observe the effects of concept modeling during the training of the LLM. Our method stands in contrast with gradient-based implicit mechanisms, which depend solely on adjustments to the LLM parameters and thus evade scrutiny. By conducting extensive experiments across both generative and discriminative language modeling tasks, we evaluate the capabilities of our proposed architecture relative to state-of-the-art LLMs of similar sizes. Additionally, we offer a qualitative examination of the interpretable read-out layer and visualize the concepts it captures. The results demonstrate the potential of our approach for effectively controlling LLM hallucinations and enhancing the alignment with human expectations.

Causal Event Graph-Guided Language-based Spatiotemporal Question Answering

Large Language Models have excelled at encoding and leveraging language patterns in large text-based corpora for various tasks, including spatiotemporal event-based question answering (QA). However, due to encoding a text-based projection of the world, they have also been shown to lack a full bodied understanding of such events, e.g., a sense of intuitive physics, and cause-and-effect relationships among events. In this work, we propose using causal event graphs (CEGs) to enhance language understanding of spatiotemporal events in language models, using a novel approach that also provides proofs for the model’s capture of the CEGs. A CEG consists of events denoted by nodes, and edges that denote cause and effect relationships among the events. We perform experimentation and evaluation of our approach for benchmark spatiotemporal QA tasks and show effective performance, both quantitative and qualitative, over state-of-the-art baseline methods.

Fully non-Gaussian Scalar-Induced Gravitational Waves

Scalar-Induced Gravitational Waves (SIGWs) represent a particular class of primordial signals which are sourced at second-order in perturbation theory whenever a scalar fluctuation of the metric is present. They form a guaranteed Stochastic Gravitational Wave Background (SGWB) that, depending on the amplification of primordial scalar fluctuations, can be detected by GW detectors. The amplitude and the frequency shape of the scalar-induced SGWB can be influenced by the statistical properties of the scalar density perturbations. In this work we study the intuitive physics behind SIGWs and we analyze the imprints of local non-Gaussianity of the primordial curvature perturbation on the GW spectrum. We consider all the relevant non-Gaussian contributions up to fifth-order in the scalar seeds without any hierarchy, and we derive the related GW energy density ΩGW(f). We perform a Fisher matrix analysis to understand to which accuracy non-Gaussianity can be constrained with the LISA detector, which will be sensitive in the milli-Hertz frequency band. We find that LISA, neglecting the impact of astrophysical foregrounds, will be able to measure the amplitude, the width and the peak of the spectrum with an accuracy up to 𝒪(10-4), while non-Gaussianity can be measured up to 𝒪(10-3). Finally, we discuss the implications of our non-Gaussianity expansion on the fraction of Primordial Black Holes.

Enhancing Robotic Perception through Synchronized Simulation and Physical Common-Sense Reasoning.

We introduce both conceptual and empirical findings arising from the amalgamation of a robotics cognitive architecture with an embedded physics simulator, aligning with the principles outlined in the intuitive physics literature. The employed robotic cognitive architecture, named CORTEX, leverages a highly efficient distributed working memory known as deep state representation. This working memory inherently encompasses a fundamental ontology, state persistency, geometric and logical relationships among elements, and tools for reading, updating, and reasoning about its contents. Our primary objective is to investigate the hypothesis that the integration of a physics simulator into the architecture streamlines the implementation of various functionalities that would otherwise necessitate extensive coding and debugging efforts. Furthermore, we categorize these enhanced functionalities into broad types based on the nature of the problems they address. These include addressing challenges related to occlusion, model-based perception, self-calibration, scene structural stability, and human activity interpretation. To demonstrate the outcomes of our experiments, we employ CoppeliaSim as the embedded simulator and both a Kinova Gen3 robotic arm and the Open-Manipulator-P as the real-world scenarios. Synchronization is maintained between the simulator and the stream of real events. Depending on the ongoing task, numerous queries are computed, and the results are projected into the working memory. Participating agents can then leverage this information to enhance overall performance.

Human perception of impact sounds suggests auditory intuitive physics

Upon hearing objects collide, humans can estimate many of the underlying physical attributes, such as the objects’ material and mass. Although the physics of sound generation are well established, the inverse problem that listeners must solve – of inferring physical parameters from sound – remains poorly understood. In this work, we show that humans leverage an understanding of acoustical physics to constrain their perceptual inferences, allowing them to disambiguate multiple object properties from a single impact sound. We derived a linear generative model of impact sounds, combining theoretical acoustics with empirically measured statistics of object resonances. We used an analysis-by-synthesis algorithm to infer mode parameters from recorded object impulse responses. We then fit distributions to these parameters, from which object impulse responses could be sampled. Perceptual experiments demonstrated that humans could judge material and mass from sound alone, even when both of the underlying objects varied. However, performance with synthetic sounds was impaired if the simulated physical regularities were altered to be unnatural. The results suggest that listeners use internal physical models to separate the acoustic contributions of the objects that interact to create sound.

Object geometry serves humans’ intuitive physics of stability

How do humans judge physical stability? A prevalent account emphasizes the mental simulation of physical events implemented by an intuitive physics engine in the mind. Here we test the extent to which the perceptual features of object geometry are sufficient for supporting judgments of falling direction. In all experiments, adults and children judged the falling direction of a tilted object and, across experiments, objects differed in the geometric features (i.e., geometric centroid, object height, base size and/or aspect ratio) relevant to the judgment. Participants’ performance was compared to computational models trained on geometric features, as well as a deep convolutional neural network (ResNet-50), none of which incorporated mental simulation. Adult and child participants’ performance was well fit by models of object geometry, particularly the geometric centroid. ResNet-50 also provided a good account of human performance. Altogether, our findings suggest that object geometry may be sufficient for judging the falling direction of tilted objects, independent of mental simulation.

Supplemental Material for Temporal Segmentation and “Look Ahead” Simulation: Physical Events Structure Visual Perception of Intuitive Physics

Supplemental Material for Temporal Segmentation and “Look Ahead” Simulation: Physical Events Structure Visual Perception of Intuitive Physics

Infants’ understanding of the causal power of agents and tools

Tools are objects that are manipulated by agents with the intention to cause an effect in the world. We show that the cognitive capacity to understand tools is present in young infants, even if these tools produce arbitrary, causally opaque effects. In experiments 1–2, we used pupillometry to show that 8-mo-old infants infer an invisible causal contact to account for the—otherwise unexplained—motion of a ball. In experiments 3, we probed 8-mo-old infants’ account of a state change event (flickering of a cube) that lies outside of the explanatory power of intuitive physics. Infants repeatedly watched an intentional agent launch a ball behind an occluder. After a short delay, a cube, positioned at the other end of the occluder began flickering. Rare unoccluded events served to probe infants’ representation of what happened behind the occluder. Infants exhibited larger pupil dilation, signaling more surprise, when the ball stopped before touching the cube, than when it contacted the cube, suggesting that infants inferred that the cause of the state change was contact between the ball and the cube. This effect was canceled in experiment 4, when an inanimate sphere replaced the intentional agent. Altogether, results suggest that, in the infants’ eyes, a ball (an inanimate object) has the power to cause an arbitrary state change, but only if it inherits this power from an intentional agent. Eight-month-olds are thus capable of representing complex event structures, involving an intentional agent causing a change with a tool.

A dedicated mental resource for intuitive physics

Countless decisions and actions in daily life draw on a mental model of the physical structure and dynamics of the world- from stepping carefully around a patch of slippery pavement to stacking delicate produce in a shopping basket. People can make fast and accurate inferences about how physical interactions will unfold, but it remains unclear whether we do so by applying a general set of physical principles across scenarios, or instead by reasoning about the physics of individual scenarios in an ad-hoc fashion. Here, we hypothesized that humans possess a dedicated and flexible mental resource for physical inference, and we tested for such a resource using a battery of fine-tuned tasks to capture individual differences in performance. Despite varying scene contents across tasks, we found that performance was highly correlated among them and well-explained by a unitary intuitive physics resource, distinct from other facets of cognition such as spatial reasoning and working memory.

Perception of 3D shape integrates intuitive physics and analysis-by-synthesis.

Many surface cues support three-dimensional shape perception, but humans can sometimes still see shape when these features are missing-such as when an object is covered with a draped cloth. Here we propose a framework for three-dimensional shape perception that explains perception in both typical and atypical cases as analysis-by-synthesis, or inference in a generative model of image formation. The model integrates intuitive physics to explain how shape can be inferred from the deformations it causes to other objects, as in cloth draping. Behavioural and computational studies comparing this account with several alternatives show that it best matches human observers (total n = 174) in both accuracy and response times, and is the only model that correlates significantly with human performance on difficult discriminations. We suggest that bottom-up deep neural network models are not fully adequate accounts of human shape perception, and point to how machine vision systems might achieve more human-like robustness.

Grounding Intuitive Physics in Perceptual Experience.

This review article explores the foundation of laypeople's understanding of the physical world rooted in perceptual experience. Beginning with a concise historical overview of the study of intuitive physics, the article presents the hypothesis that laypeople possess accurate internalized representations of physical laws. A key aspect of this hypothesis is the contention that correct representations of physical laws emerge in ecological experimental conditions, where the scenario being examined resembles everyday life experiences. The article critically examines empirical evidence both supporting and challenging this claim, revealing that despite everyday-life-like conditions, fundamental misconceptions often persist. Many of these misconceptions can be attributed to a domain-general heuristic that arises from the overgeneralization of perceptual-motor experiences with physical objects. To conclude, the article delves into ongoing controversies and highlights promising future avenues in the field of intuitive physics, including action-judgment dissociations, insights from developmental psychology, and computational models integrating artificial intelligence.

Do striking biases in mass inference reflect a flawed mental model of physics?

To engage with the physical world, we rely on our intuitive sense of how objects will behave when we act on them or they interact with each other. Objects' latent properties such as mass and hardness determine how their physical interactions will unfold, and people have a keen ability to infer these latent properties by observing physical events. For example, we can precisely discriminate the relative masses of two objects when we see them collide. However, such inferences are sometimes subject to marked biases. When inferring mass from an observed collision, people consistently overestimate the mass of an incoming object that strikes a stationary one. Why? A number of plausible accounts have been put forward, variously arguing that the bias arises from rule-based reasoning, oversimplified stimuli, or noisy perceptual estimates of the scene dynamics. The implications of these views stand in stark contrast to one another: systematic biases may reveal a fundamental deficiency in the mental model of physical behavior, or they may be an expected consequence of reasoning over imperfect information. Here, we investigated all three accounts within a unified paradigm, presenting videos of real-world bowling ball collisions. We found that using richly detailed stimuli did not eliminate biases in mass inference. However, individual differences in the biases were task-specific and well-explained by noisy perceptual estimates rather than oversimplified physical inference mechanisms. Our findings collectively point toward an intuitive physics system that implements Newtonian principles but is subject to the quality of the information it operates on. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Intuitive physics guides visual tracking and working memory: The dynamics of neural processing in expectation violation

Intuitive physics guides visual tracking and working memory: The dynamics of neural processing in expectation violation

Linking Intuitive Physics to Social Cognitive Attributions

Linking Intuitive Physics to Social Cognitive Attributions