STRIPS

STRIPS
Name	STRIPS
Caption	STRIPS planning formalism
Introduced	1971
Inventor	Richard E. Fikes; Nils J. Nilsson
Discipline	Artificial intelligence
Related	Planning problem; Automated planning; Heuristic search; Classical planning

STRIPS is a formal language and framework for automated planning in Artificial intelligence. Developed to model action, state, and goal specifications, it provides a concise representation used by planners and researchers across Stanford University, SRI International, DARPA, Massachusetts Institute of Technology, and corporate labs. STRIPS influenced subsequent systems and formalisms in Robotics, Cognitive science, Computer science, Operations research, and Control theory.

History

STRIPS was introduced in 1971 by Richard E. Fikes and Nils J. Nilsson at Stanford Research Institute (now SRI International) and presented in work connected to projects supported by DARPA. Early demonstrations linked STRIPS-based planners to robotics efforts at Stanford University and vehicles of interest to NASA research. The formalism built on antecedents in symbolic reasoning from John McCarthy, Marvin Minsky, and researchers at MIT, and it quickly spread through communities at Carnegie Mellon University, University of California, Berkeley, and University of Edinburgh. STRIPS framed much of the 1970s and 1980s discourse in planning, shaping competitions such as the International Conference on Automated Planning and Scheduling and influencing languages later standardized by the International Planning Competition and the development of PDDL.

Formal definition

STRIPS defines a planning problem by an initial state, a goal specification, and a set of actions. Each action has an add list and a delete list, together with preconditions expressed as conjunctive predicates over a finite set of ground atoms drawn from a domain vocabulary. The formal semantics relate to state-transition systems studied in Set theory and Logic programming and are expressed using operators akin to those in transitional models in Tarski-style semantics. The goal test is typically a conjunction of ground literals; plan validity is proven through state reachability under deterministic transitions. The formal model connects to complexity results in Stephen Cook-style computational theory and to decision problems analyzed by researchers at Princeton University and Bell Labs.

STRIPS language and representation

The STRIPS representation uses predicates, constants, and variables to encode domain facts and object properties. Domains are often expressed in files associated with planners developed at Stanford University labs and adapted in languages like PDDL used in International Planning Competition benchmarks. Typical predicates in classical STRIPS encodings mirror examples from robotics tested at NASA Ames Research Center and industrial demonstrations at General Electric and IBM Research. Action schemata specify parameter lists, preconditions, add lists, and delete lists; the representation is compatible with substitutions and unification techniques formalized by researchers at University of Edinburgh and University of Massachusetts Amherst. STRIPS limitations motivated extensions such as conditional effects, quantified preconditions, and temporal constructs adopted by planners in projects at Carnegie Mellon University and University of Tokyo.

Planning algorithm and extensions

Classical STRIPS planning algorithms include forward state-space search, backward goal regression, and plan-space search; these strategies were explored in systems from SRI International, Stanford University, and IBM Research. Heuristic search methods such as A* and best-first search were adapted to STRIPS by researchers at University of California, Berkeley and University of Illinois Urbana–Champaign, while landmark and delete-relaxation heuristics trace lineage to work at University of Strathclyde and University of Freiburg. Extensions to the basic STRIPS model incorporate conditional effects, nondeterministic actions, and partial observability developed in collaborations with teams at MIT, University College London, and University of Washington. Integration with temporal planning and resource reasoning appears in frameworks produced at Cornell University and ETH Zurich.

Complexity and computational properties

Decision and optimization problems for STRIPS relate to classical complexity classes studied by Alan Turing-inspired theory and formalized in texts by Michael Sipser and Christos Papadimitriou. Plan existence for general propositional STRIPS is PSPACE-complete; bounded-length plans yield NP-complete formulations under encodings used in competitions such as the International Joint Conference on Artificial Intelligence. Complexity analyses influenced algorithmic advances at Princeton University and lower-bound reductions often cite canonical problems like those studied by Stephen Cook and Leonid Levin. Practical heuristics and compilation schemes trade off worst-case hardness against empirical performance, a theme explored at University of Toronto and University of California, Los Angeles.

Applications and implementations

STRIPS-style models underpin classical planners and have been embedded in robotic architectures at NASA, industrial scheduling systems at Siemens, and autonomous agents developed at Lockheed Martin and DARPA programs. Implementations include early planners from SRI International and later engines such as GRAPHPLAN-derived systems, heuristic planners used by teams at University of Maryland and PDDL-based planners submitted to the International Planning Competition. STRIPS representations inform automated workflow systems in Microsoft Research and diagnostic planning tools in CERN experiments. Contemporary use often appears in hybrid systems combining STRIPS-like symbolic planning with learning components developed at Google DeepMind and OpenAI.

Category:Automated planning