Compass: Devices, Models and Applications

• Compass is a family of systems defined by directional sensing and structured choice, spanning magnetic devices, geometric constructions, educational initiatives, and high-energy experiments.
• Research methodologies include torque-based alignment, inversion geometry, and multimodal pretraining that achieve milli radian accuracy and precise positioning in various applications.
• Compass approaches drive innovation in VR navigation, particle spectroscopy, and autonomous systems by transforming complex environments into oriented, actionable decision frameworks.

In the literature represented here, compass denotes a family of devices, models, and research systems organized around orientation, direction finding, or structured choice. The term includes the magnetic field-direction indicator, the geometric construction instrument of compass-only constructions, the Berkeley-based Compass Project in physics education, the CERN fixed-target experiment COMPASS, and multiple acronymic systems in autonomous navigation, localization, privacy, and medical uncertainty quantification (Wojtsekhowski, 2024, Margaryan et al., 2014, Roth et al., 2012, Schill, 2011, Ma et al., 2022).

1. Magnetic direction finding and resonant behavior

A magnetic compass, in the strict physical sense used in the cited work, aligns through the torque $\vec B \times \vec M$ between an external magnetic field and the magnetic moment of the sensing element; when $\vec B$ and $\vec M$ are parallel, the torque vanishes. The 2024 “spinning Hall probe magnetic compass” replaces the torque-driven needle with a rotating Hall probe whose plane is mounted approximately parallel to the axis of rotation, so that the magnetic field component transverse to the rotation axis becomes an alternating Hall signal proportional to $\vec B \times \vec n_A$. Direction finding is then performed by nulling that synchronous AC component: when the alternating signal reaches zero, the spin axis is aligned with the field. The prototype did not require calibration, was described as free of drift problems, and achieved magnetic-field direction accuracy on the level of one milli radian in a $25\ \mathrm{G}$ polarized ${}^3\mathrm{He}$ target experiment; the paper also reports a $1\ \mathrm{mG}$ residual response under $\mu$-metal shielding, corresponding to about $0.04\ \mathrm{mrad}$ systematic angular accuracy for that field strength (Wojtsekhowski, 2024).

The same literature also uses a hand-held compass as a deliberately simple analog of magnetic resonance. In that experiment, a compass is placed at the center of a coil and near a permanent magnet so that the static field and the alternating drive field are perpendicular. The field near a typical permanent dipole magnet is stated to be on the order of $\sim 5\times10^{-3}\ \mathrm{Tesla}$, much larger than the Earth’s field $\vec B$0, so the nearby magnet sets the relevant equilibrium direction. Under the small-angle approximation, the needle obeys the driven-oscillator equation

$\vec B$1

with resonance frequency

$\vec B$2

At fixed drive frequency, changing the magnet distance tunes $\vec B$3 and produces the visible “sweet spot” at resonance. The authors are explicit that this is not a literal model of quantum spin precession: the compass needle oscillates mechanically in a plane, whereas real spin magnetic moments precess (Cookson et al., 2018).

2. Unconventional compass embodiments and interface cues

The term also appears in nanoscale materials science in a form that reverses ordinary intuition. The “nanoscale magnetic compasses” are ferromagnetic iron-carbide particles encapsulated in a pair of parallel graphitic carbon needles. Electron holography shows magnetic fields confined to the vicinity of the bicone-shaped iron-carbide core, which consists of a few ferromagnetic domains, while the carbon arms show no magnetic ordering. Their defining feature is that the structural needle axis and the magnetic easy axis are orthogonal: the visible needle-like body aligns perpendicular to an applied field because the magnetic core magnetizes most easily normal to the carbon-needle direction. Bulk measurements show anisotropic permeability with the easy axis normal to the needle direction, and the proposed applications include electromagnetic wave absorbent materials and magnetorheological fluids (Shiozawa et al., 2017).

A different, nonmaterial sense appears in immersive virtual reality, where a compass is a navigation aid rather than a physical field sensor. In the cited room-scale VR study, the compass was implemented as a horizontal compass bar at the bottom of the field of view indicating target heading relative to current orientation. It provided “goal bearing only,” without route feasibility or allocentric layout. In a $\vec B$4 maze task with 42 participants and 1008 trials under reduced visibility, time pressure, and forced route replanning, arrow guidance produced the strongest navigation performance, minimap guidance was intermediate, and compass cues performed worst. Estimated marginal means for the composite navigation score were Arrow $\vec B$5, Minimap $\vec B$6, and Compass $\vec B$7; compass users also showed the highest interface dwell time and higher stress than the other aided conditions. The result is not that compass cues are universally ineffective, but that in this demanding embodied-locomotion regime a bearing-only representation imposed more cognitive translation than an actionable egocentric arrow (Varshney et al., 18 Mar 2026).

3. Compass-only construction in geometry

In classical geometry, the compass is the construction instrument whose scope is tested by the Mohr–Mascheroni-type theorem treated in “Compass Constructions.” The manuscript proves that every point construction achievable by ruler and compass can also be achieved with a compass alone. It reduces the entire problem to two tasks from four given points $\vec B$8: constructing the intersection of the two lines $\vec B$9 and $\vec M$0, and constructing the intersection of line $\vec M$1 with the circle centered at $\vec M$2 and passing through $\vec M$3. Intersections of two circles are treated as immediate, because drawing circles is exactly what the compass does (Margaryan et al., 2014).

The proof strategy combines algebraic and inversion-based geometry. The plane is identified with $\vec M$4, the set of compass-constructible points is shown to be a subring closed under conjugation, and midpoint construction is obtained so that circles with given diameters and feet of perpendiculars become compass-constructible. Inversion then becomes the principal mechanism: a line not through the inversion center is mapped to a circle through that center, and line-line or line-circle problems are therefore converted into circle-circle intersections. The manuscript gives explicit procedures for constructing inverse points with respect to a circle, for recovering the intersection $\vec M$5 by inverting the second intersection of two auxiliary circles, and for handling line-circle intersection both when the circle center lies off the line and when it lies on the line (Margaryan et al., 2014).

4. Compass as an educational community

The Compass Project at the University of California, Berkeley is a student-created, student-run, and evolving physics-education community founded in 2006 by physics graduate students Angie Little, Hal Haggard, and Badr Albanna. Its stated goals are to improve undergraduate physics education, provide professional-development opportunities, and increase retention of students, especially those from populations typically underrepresented in the physical sciences. The paper stresses that Compass is not a remedial program; rather, it aims to “help empower students to find their way forward in college and life by weaving together a community,” through intellectually challenging science, collaborative learning, mentoring, and shared leadership (Roth et al., 2012).

Its first summer program took place in August 2007 and brought 11 incoming freshmen together for two weeks before the fall semester. The initial curriculum centered on the question, “What can earthquakes tell us about the interior of the Earth?”, and students used wave mechanics, mass-spring models, and earthquake data to infer the size of Earth’s liquid core. This inquiry-based, model-centered, collaborative style remained characteristic of Compass, with later summer themes including wind turbines, non-Newtonian fluids, and “levitating” Slinkies. The program subsequently expanded to include a mentoring system pairing undergraduates with graduate-student mentors, a research lecture series, office hours and academic support sessions, social events, a fall course on physical modeling added in 2009, a spring course on measurement and error analysis added in 2012, and a pilot course for incoming transfer students (Roth et al., 2012).

Admissions are not based on standardized test scores or GPA, but on short essays about academic interests, experiences with diversity, enthusiasm for science, and desire to contribute to a community. Between 2007 and 2012, six summer programs served 88 students; among those participants, 45% were female, 19% were first-generation college students, 26% were Chicano/Latino, 5% were African American, and 1% were Native American. Among the 26 students who entered Berkeley through Compass in 2007 and 2008, 58% had completed a science or engineering degree at the time of reporting, compared with a national figure of 38% of entering science-and-engineering freshmen completing such a degree within six years. The authors present this not as a causal estimate but as suggestive evidence. By the time of the paper, Compass was in its seventh year and had received the American Physical Society’s Award for Improving Undergraduate Physics Education (Roth et al., 2012).

5. COMPASS at CERN

In uppercase form, COMPASS denotes the CERN fixed-target experiment described as the COmmon Muon and Proton Apparatus for Structure and Spectroscopy. It operates at the CERN SPS as a multi-purpose fixed-target program in nucleon structure and hadron spectroscopy. In the transverse-spin program, COMPASS uses a $\vec M$6 muon beam and transversely polarized $\vec M$7 or $\vec M$8 targets to access transversity and transverse-momentum-dependent structure through semi-inclusive deep-inelastic scattering. The experiment measured Collins, Sivers, and two-hadron asymmetries; on the proton target, the Collins asymmetry for identified pions is small up to about $\vec M$9 and then reaches roughly 10% in higher-$\vec B \times \vec n_A$0 bins, with opposite signs for positive and negative pions, while the Sivers asymmetry is positive for positive pions and compatible with zero for negative pions. Earlier deuteron asymmetries were generally compatible with zero, and the standard interpretation given is cancellation between $\vec B \times \vec n_A$1- and $\vec B \times \vec n_A$2-quark contributions in an isoscalar target (Schill, 2011).

COMPASS also developed a dedicated generalized parton distribution program. The cited GPD study emphasizes the unique $\vec B \times \vec n_A$3 to $\vec B \times \vec n_A$4 region accessible with the $\vec B \times \vec n_A$5 polarized muon beam of either charge, and describes the transition from solid-state polarized targets to the COMPASS-II setup with a $\vec B \times \vec n_A$6 liquid-hydrogen target, the CAMERA recoil proton detector, and ECAL0. In exclusive $\vec B \times \vec n_A$7 muoproduction on transversely polarized protons, the asymmetry

$\vec B \times \vec n_A$8

was found to be nonzero, and the paper interprets this as the first experimental evidence in hard exclusive $\vec B \times \vec n_A$9 leptoproduction for the existence of the transverse chiral-odd GPD $25\ \mathrm{G}$0 (Sandacz, 2015).

The hadron-spectroscopy program uses $25\ \mathrm{G}$1 hadron beams on hydrogen and nuclear targets to study diffractive and central production. Earlier reports emphasized the need to control Deck-like and double-Regge backgrounds when interpreting exotic-wave signals, but the 2019 review concludes that COMPASS data provide solid evidence for the manifestly exotic $25\ \mathrm{G}$2, whose $25\ \mathrm{G}$3 quantum numbers are forbidden for a quark-model state, and for the $25\ \mathrm{G}$4, which does not fit naturally into the quark-model spectrum. The same review describes the most comprehensive analyses to date of several isovector light-meson channels and notes that, by isolating quasi-real photon exchange on heavy nuclei, COMPASS measured the radiative widths of the $25\ \mathrm{G}$5 and, for the first time, the $25\ \mathrm{G}$6 (Grube, 2013, Ketzer et al., 2019).

The acronym also extends to major detector subsystems. The PixelGEM central tracking system, installed in spring 2008, consists of five triple-GEM detectors with a hybrid readout combining a $25\ \mathrm{G}$7 central pixel region of $25\ \mathrm{G}$8 pixels and a surrounding 2-D strip readout, all in a $25\ \mathrm{G}$9 active area and a total of 2048 channels. Its total material budget is ${}^3\mathrm{He}$0, and beam tests reported plateau efficiencies of ${}^3\mathrm{He}$1 for low-intensity muon beams and ${}^3\mathrm{He}$2 for high-intensity muon beams, spatial resolutions of ${}^3\mathrm{He}$3 and ${}^3\mathrm{He}$4 in the pixel region, and time resolution below ${}^3\mathrm{He}$5 (Austregesilo et al., 2013). Electromagnetic calorimetry is similarly central: ECAL1 measures ${}^3\mathrm{He}$6 with 1492 channels, ECAL2 measures ${}^3\mathrm{He}$7 with 3072 channels, and the planned ECAL0 was introduced for large-angle DVCS photons. Under the future three-calorimeter geometry discussed in the calorimetry note, COMPASS could detect approximately ${}^3\mathrm{He}$8 of all produced DVCS photons, with contributions of 43% from ECAL1, 23% from ECAL2, and 22% from ECAL0 (Nerling, 2010).

6. Extended and acronymic usages in modern research

The uppercase form COMPASS has been repeatedly reused as a domain-specific acronym for systems that infer direction, structure, or admissible action in a latent, geometric, or operational space.

System	Domain	Core mechanism
“COMPASS: Contrastive Multimodal Pretraining for Autonomous Systems” (Ma et al., 2022)	Autonomous systems	A multimodal graph maps RGB, depth, and optical flow into two factorized latent spaces: a “current state space” and a “motion pattern space.”
“COMPASS: COmpact Multi-channel Prior-map And Scene Signature for Floor-Plan-Based Visual Localization” (Shaheer et al., 28 Apr 2026)	Indoor robot localization	A ${}^3\mathrm{He}$9 radial descriptor encodes normalized range, structural hit type, range gradient, inverse range, and local range variance from floor plans and dual fisheye images.
“COMPASS: VLBI Beacons In Support of Lunar Science and Exploration” (Eubanks, 2020)	Cislunar positioning, navigation, and timing	Coherent ultra-wideband spacecraft beacons interoperable with VLBI networks are used for differential phase-referenced positioning.
“CompassNav: Steering From Path Imitation To Decision Understanding In Navigation” (Li et al., 11 Oct 2025)	LVLM navigation	A 22k-trajectory dataset and a gap-aware hybrid reward shift training from single-path imitation to comparison of all feasible actions.
“A-COMPASS: Formal Foundations for Anonymity Analysis in Microdata” (Tagliavia et al., 18 Jun 2026)	Formal privacy verification	An extension of COMPASS to one-record-one-person microdata with syntax, semantics, anonymization actions, and proofs of determinism and compositionality.
“COMPASS: Robust Feature Conformal Prediction for Medical Segmentation Metrics” (Cheung et al., 26 Sep 2025)	Medical uncertainty quantification	Conformal intervals for segmentation-derived scalar metrics are generated by perturbing intermediate features along low-dimensional subspaces sensitive to the target metric.

Across these systems, the term consistently signals guided orientation rather than mere naming. In multimodal autonomy, COMPASS learns separate latent factors for instantaneous scene state and temporal dynamics, and improves generalization to unseen racing environments and real-world visual odometry data (Ma et al., 2022). In floor-plan localization, the same name denotes a semantic-geometric radial descriptor whose proof-of-concept cross-modal matching on the Hilti-Trimble SLAM Challenge 2026 data produced a correlation peak of $1\ \mathrm{mG}$0 at $1\ \mathrm{mG}$1 shift (Shaheer et al., 28 Apr 2026). In cislunar navigation, COMPASS becomes Combined Observational Methods for Positional Awareness in the Solar System: the paper argues that multi-baseline phase-referenced VLBI with coherent ultra-wideband beacons can provide picosecond-level phase-delay determination, sub-meter transverse positioning, and meter-level lunar orbit determination, with a detectability estimate of SNR 10 in 1 second for a 1 mW beacon at lunar distance under stated assumptions (Eubanks, 2020).

The same orientation metaphor is explicit in machine navigation, privacy, and medical AI. CompassNav argues that navigation agents should learn Decision Understanding rather than replicate a single expert route, and reports that its 7B agent reaches an average SR/SPL of $1\ \mathrm{mG}$2, outperforming the listed proprietary-model baselines on the reported goal-navigation benchmarks (Li et al., 11 Oct 2025). A-COMPASS shifts from group-preprocessed tables to ordinary microdata, adds REPLACE and COUNT DISTINCT, and proves determinism and sequential compositionality while supporting direct verification of $1\ \mathrm{mG}$3-anonymity and $1\ \mathrm{mG}$4-diversity (Tagliavia et al., 18 Jun 2026). In medical imaging, COMPASS calibrates prediction intervals for downstream segmentation metrics in representation space rather than on the final scalar alone; the paper proves valid marginal coverage under exchangeability and nestedness assumptions and reports substantially tighter intervals than scalar-output conformal baselines on four area-estimation tasks (Cheung et al., 26 Sep 2025).

A plausible unifying implication is that, across physics, geometry, education, particle experiments, robotics, privacy, and medical inference, compass functions less as a single object than as a recurring technical pattern: a mechanism for converting a complex environment into an oriented decision, alignment, or construction problem.