pyspect

Compile temporal-logic specs into reachability programs.

pyspect lets you write specifications once, then realize them against interchangeable backends (e.g., Hamilton–Jacobi level sets or Hybrid Zonotopes) via Temporal Logic Trees (TLTs). The toolbox performs interface + approximation checks so your output set remains sound.

TL;DR: You express what to verify in logic; implementations decide how to compute the sets. pyspect bridges the two with TLTs and small, pluggable operator primitives. See paper "pyspect: An Extensible Toolbox for Automatic Construction of Temporal Logic Trees via Reachability Analysis."

Why pyspect

• Decouple logic, set semantics, and numerics: write specs once, compare multiple reachability methods side-by-side.
• Method-agnostic and easily extensible: currently supports HJ (level sets) and HZ-based (hybrid zonotopes) reachability within the same interface.
• Correct by construction: static checks for approximation direction (over/under) and backend capability before any heavy computation, avoiding invalid evaluation of the spec.

Getting started

Read our docs here!

Installation

pip install pyspect

# Example to install with implementation-specific dependencies (Optional)
pip install pyspect[hj_reachability]

# From source
python -m venv .venv && source .venv/bin/activate
pip install -e ".[hj_reachability]"

Program

from pyspect.logics import *
from pyspect.tlt import TLT, ContLTL
from pyspect.impls.hj_reachability import HJImpl

# 1) Pick primitives/fragment
TLT.select(ContLTL) # Continuous-time LTL

# 2) Write the spec once
phi = UNTIL(AND(NOT('D'), 'corridor'), 'goal')   # task: avoid D, stay in corridor, then reach goal

# 3) Bind propositions later via set builders
tlt = TLT(phi, where={
    'D': Union(BoundedSet(...), BoundedSet(...)),
    'corridor': BoundedSet(...),
    'goal': HalfSpaceSet(...),
})

# 4) Realize on a backend
impl = HJImpl(...)      # Each implementation can have their own settings
Phi = tlt.realize(impl) # The satisfaction set in the backend’s representation

Cite

If you use pyspect in academic work, please cite:

TBA

Paper Examples

For paper examples, checkout branches:

• cdc24: Intersection & Parking
• cdc25: Double Integrator with HJ & HZ