PREGO: An Action Language for Belief-Based Cognitive Robotics in Continuous Domains

Authors: Vaishak Belle, Hector Levesque

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We consider the empirical behavior of prego in this section, where we measure the CPU time in milliseconds (ms) on non-trivial projection tasks, which is then contrasted with classical regression in a manner explained shortly. For the experimental setup, we imagine a planner to have generated a number of plans of increasing length. For our purpose, these plans are randomly generated action sequences, possibly involving combinations of noisy and noise-free actions and sensors, and thus, are representative of the plan search space. All experiments were run wrt the simple robot domain, on Mac OS X 10.8 using a system with a 2.4 GHz Intel Core 2 Duo processor, 4 GB RAM, and racket v5.3.6.
Researcher Affiliation Academia Vaishak Belle and Hector J. Levesque Dept. of Computer Science University of Toronto Toronto, Ontario M5S 3H5, Canada {vaishak, hector}@cs.toronto.edu
Pseudocode Yes Definition 1: Given a BAT D, a situation-suppressed expression e, and an action sequence σ, we define R[e, σ] as a situation-suppressed formula e as follows: 1. If e is a fluent: if σ = ϵ (is empty), then e = e; if σ = σ a then e = R(SSAe(a), σ ). 2. If e is a number, constant or variable, then e = e. 3. If e is Bel(φ, now) then f INIT G[φ, σ] where G is an operator for obtaining a (mathematical) expression from the belief argument φ, defined below. 4. Else e is (e1 e2 . . . en) and e = (R[e1, σ] R[e2, σ] . . . R[en, σ]) where is any mathematical operator over expressions, such as , , =, +, If, N, etc. ... Definition 2: Let D and σ be as above. Given any situationsuppressed fluent formula φ, we define G[φ, σ] to be a situation-suppressed expression e as follows: 1. If σ = ϵ, then e = If φ Then 1 Else 0. 2. Else, σ = σ α(t); let α(t ) = alt(α(t), u) and u R[Lα(t ), σ ] G[R[φ, α(t )], σ ].
Open Source Code No The paper mentions that "prego is fully implemented" and "The system is realized in the racket dialect of the scheme family (racket-lang.org)" but does not provide an explicit statement or link for the open-source availability of the prego system's code.
Open Datasets No The paper describes a simulated robot domain with an initial uniform distribution for the fluent 'h' on [2,12], but it does not use or provide concrete access information for a publicly available or open dataset.
Dataset Splits No The paper describes experiments run on generated action sequences within a simulated robot domain but does not provide specific dataset split information (e.g., percentages, sample counts, or predefined splits) for training, validation, or testing.
Hardware Specification Yes All experiments were run wrt the simple robot domain, on Mac OS X 10.8 using a system with a 2.4 GHz Intel Core 2 Duo processor, 4 GB RAM, and racket v5.3.6.
Software Dependencies Yes The system is realized in the racket dialect of the scheme family (racket-lang.org)." and "All experiments were run wrt the simple robot domain... and racket v5.3.6.
Experiment Setup Yes The prego language is a simple representation language with a LISP-like syntax. A domain in prego is modeled as a basic action theory (or BAT) made up of the following five expressions (which we will illustrate immediately below): 1. (define-fluents fluent fluent ...) 2. (define-ini-p-expr expr) 3. (define-ss-exprs fluent act expr act expr . . .) 4. (define-l-exprs act expr act expr . . .) 5. (define-alts act altfn act altfn . . .) and "For the experimental setup, we imagine a planner to have generated a number of plans of increasing length. For our purpose, these plans are randomly generated action sequences, possibly involving combinations of noisy and noise-free actions and sensors, and thus, are representative of the plan search space.