MIT PNAS paper: humans infer physical properties from impact sounds using internalized object acoustics

MIT's Josh McDermott lab released a new paper on November 10 on how humans perceive the physical properties of objects from impact sounds. Authored by Vin Agarwal (first author) with James Traer, Schwartz, and McDermott, and published in PNAS 2026, it is titled Intuitive knowledge of object acoustics enables perception of physical variables from impact sounds. The work caps roughly a decade of research in the lab, approaching auditory "intuitive physics" as the entry point.

Confirmed

• Core claim: Inferring physical variables such as mass, material, size, shape, and damping from impact sounds is essentially an ill-posed scene-analysis problem, which humans solve using internalized real-world statistical knowledge of "how objects sound."
• Methods included measuring the acoustic properties of hundreds of everyday objects (glass, ceramics, etc.) to build a real-world generative model; accompanying figures show how impact force and an object's impulse response combine into an impact sound when a ball strikes a marble slab.
• The authors use an example task to illustrate the difficulty: judging physical differences, such as relative mass, between two balls dropped on different surfaces using sound alone.
• Experimental evidence: Beyond mass discrimination, the team also had subjects judge from sound alone whether an object was held in a hand (damped). Both experiments converged—human performance suffered most under stimuli least consistent with the real-world generative model, indicating the brain uses statistical priors over real objects to constrain physical inference.

Why it matters

• The study reveals the auditory system's "intuitive physics" mechanism: rather than recovering physical quantities directly from acoustic signals, the brain performs joint inference leveraging internalized knowledge of real-object acoustics, turning an underdetermined problem into a solvable one.
• This informs our understanding of the computational principles of human perception and offers human-like priors for machine hearing and physical inference systems.

2026-11-10 ~ 2026-11-10 · 8 related posts

Full story(5 episodes)→

• Episode 1: MIT PNAS paper: humans infer physical properties from impact sounds using internalized object acoustics(2026-11-10, 8 posts)
• Episode 2: MIT measures impulse responses of hundreds of everyday objects(2026-11-10, 2 posts)
• Episode 3: Generative model samples realistic impact sounds from measured object acoustics(2026-11-10, 2 posts)
• Episode 4: Human hearing relies on real-world sound statistics for physical inference, study finds(2026-11-10, 2 posts)
• Episode 5: MIT releases decade-long project's impact-sound data and models(2026-11-10, 2 posts)

• MIT's 10-year study uses impact sounds to reveal how humans perceive physics by ear — JoshHMcDermott · 2026-11-10
• [source] MIT's McDermott lab: humans infer physical properties from sound using internalized statistics of how objects make sound — JoshHMcDermott · 2026-11-10
• [source] MIT lab's PNAS paper: humans use internalized object acoustics to parse impact sounds — JoshHMcDermott · 2026-11-10
• [source] MIT lab studies intuitive physics of sound: how humans infer object properties from impact sounds — JoshHMcDermott · 2026-11-10
• MIT lab paper uses impact-sound perception to probe audio-based intuitive physics — JoshHMcDermott · 2026-11-10
• New McDermott lab study: how humans judge ball mass from impact sounds alone — JoshHMcDermott · 2026-11-10
• McDermott lab: humans use real-world acoustic priors to constrain physical inference — JoshHMcDermott · 2026-11-10
• Humans solve ill-posed physics-from-sound via joint inference of object variables, MIT paper shows — JoshHMcDermott · 2026-11-10