Skip to content

Scientific domain glossary#

The framework's abstractions map directly onto behavioral-neuroscience concepts. This glossary defines the domain terms — most sharply exercised by the VR Foraging paradigm, whose white paper is the canonical source. Terminology here follows that project's style guide (e.g. "stop and harvest / choice", never "poke" or "response").

Framework-level terms#

  • Rig — the physical apparatus: Harp devices + COM ports, olfactometer channels/odorants, treadmill and water-valve calibrations, cameras, motorized spout, display. Modeled by Rig.
  • Task — what the animal experiences: environment structure, patches, reward dynamics, virtual sites, operation control, adaptive updaters. Modeled by Task.
  • Session — one run: subject, date, experimenter, output paths. Modeled by Session.

VR-Foraging paradigm terms#

The paradigm: a thirsty, head-fixed mouse runs on a treadmill; its locomotion drives forward motion through a rendered linear virtual corridor tiled into typed sites. Patches are marked by odor cues; harvesting depletes a patch, so the animal must decide when to leave and pay a travel cost to reach a fresher one.

  • Virtual site — a tile of the corridor. Types: InterPatch (travel segment = the travel cost), InterSite (spacing within a patch), RewardSite (odor-cued harvest location), PostPatch (optional post-patch segment, e.g. added friction). Lengths are drawn from parameterized distributions (typically truncated exponentials).
  • Patch — a group of sites sharing: an odor specification (a mixture vector of up to 3 odor-channel concentrations = the identity cue), a reward specification (amount × probability × available count, plus optional operant logic), a visual stimuli specification, a per-site friction coefficient (rendered by the treadmill brake), reward dynamics (chained reward functions implementing depletion/replenishment), and patch terminators (end conditions).
  • Odor site / RewardSite — where harvest attempts happen; odor-cued.
  • Reward — water; delivered via GiveReward / ForceGiveReward events.
  • Choice vs. rejection — at an odor-cued reward site the animal either stops and harvests (a choice) or runs through (a rejection). Operant logic may require the animal to hold velocity below a threshold for a stop_duration to count as a stop.
  • Trial — not a native unit of this paradigm; the atomic unit is the patch / reward-site / choice. "Trial" appears only in downstream analysis (dashboard trial tables built via an NWB TrialTableProcessor).
  • Environment model — a Markov environment (patches + transition matrix
  • first-state occupancy) or a sequence environment (ordered/sampled patch indices), grouped into blocks with end-conditions that enable within-session contingency changes.

Task-DSL primitives#

Composable grammar pieces (VR Foraging glossary): RewardFunction (Patch / Outside / OnThisPatchEntry / Persistent), PatchUpdateFunction (ClampedRate / Multiplicative / Saturating / SetValue / LookupTable / Ctcm = continuous-time Markov), PatchTerminator, VirtualSiteGenerator, OperantLogic, NumericalUpdater, and OperationControl (Odor / Position / Audio).

Scientific motivation (why the paradigm)#

Foraging is ethologically natural and quantitatively tractable (optimal foraging theory / Marginal Value Theorem; reinforcement learning). Two headline claims motivate the design: mice are sensitive to reward statistics (they track depletion and adjust leaving decisions), and the task evokes cognition via stimulus–action dissociation (the same odor cue can demand opposite actions depending on internal state), forcing reliance on latent internal variables in a way a two-alternative choice task cannot.

Citations#

  1. vr-foraging-white-paper WHITEPAPER_DRAFT.md (paradigm, glossary Appendix A)
  2. vr-foraging-white-paper docs/context/style-guide.md (terminology table)