REACT! An Interactive Tool for
Hybrid Planning in Robotics
Zeynep Dogmus, Esra Erdem, and Volkan Patoglu
Sabancı University, ˙Istanbul, Turkey
{zeynepdogmus,esraerdem,vpatoglu}@sabanciuniv.net
Abstract. We present REACT!, an interactive tool for high-level reasoning for
cognitive robotic applications. REACT! enables robotic researchers to describe
robots’ actions and change in dynamic domains, without having to know about
the syntactic and semantic details of the underlying formalism in advance, and
solve planning problems using state-of-the-art automated reasoners, without hav-
ing to learn about their input/output language or usage. In particular, REACT! can
be used to represent sophisticated dynamic domains that feature concurrency, in-
direct effects of actions, and state/transition constraints. It allows for embedding
externally deﬁned calculations (e.g., checking for collision-free continuous tra-
jectories) into representations of hybrid domains that require a tight integration
of (discrete) high-level reasoning with (continuous) geometric reasoning. RE-
ACT! also enables users to solve planning problems that involve complex goals.
Such variety of utilities are useful for robotic researchers to work on interesting
and challenging domains, ranging from service robotics to cognitive factories.
REACT! provides sample formalizations of some action domains (e.g., multi-
agent path planning, Tower of Hanoi), as well as dynamic simulations of plans
computed by a state-of-the-art automated reasoner (e.g., a SAT solver or an ASP
solver).
Keywords: Reasoning about actions and change, AI planning, robotics
1
Introduction
As the robotics technology makes its transition from repetitive tasks in highly-structured
industrial settings to loosely deﬁned tasks in unstructured human environments, sub-
stantial new challenges are encountered. For instance, in order to be able to deploy
robotic assistants in our society, these systems are expected to robustly deal with high
complexity and wide variability of their surroundings to perform typical everyday tasks
without sacriﬁcing safety. Moreover, whenever more than one robotic agent is available
in a domain, these robots are expected to collaborate with each other intelligently to
share common tasks/resources. The complexity of the tasks and the variability of envi-
ronment place high demands on the robots’ intelligence and autonomy. Consequently,
there exists a pressing need to furnish robotic systems with high-level cognitive capa-
bilities.
The multidisciplinary ﬁeld of robotics is diverse and the technical background of
robotics researchers are highly heterogenous. Even though artiﬁcial intelligence (AI)
arXiv:1307.7494v1  [cs.AI]  29 Jul 2013
planning and reasoning about actions and change have been studied for decades in the
ﬁeld of computer science, leading to various action description languages, computa-
tional methods and automated reasoners, utilization of these outcomes in robotic sys-
tems has only recently gained momentum, thanks to increasing demands from new chal-
lenging domains, like service robotics applications. However, as many of the robotics
researchers trained in diverse engineering aspects of robotics are not familiar with these
logic-based formalisms, underlying theoretical AI concepts, and the state-of-the-art au-
tomated reasoners, it is still a challenge to integrate high-level automated reasoning
methods in robotics applications.
In this paper, we introduce an interactive tool, called REACT!, to fulﬁll this need in
robotics. With REACT!, robotics researchers can describe robots’ actions and change
in a dynamic domain, without having to know about the syntactic and semantic details
of the underlying formalism in advance. Such dynamic domains may be quite sophisti-
cated, allowing concurrency, indirect effects of actions, and state/transition constraints.
They can also solve planning problems, which may involve temporal complex goals, us-
ing a state-of-the-art automated reasoner, without having to know about its input/output
language or usage. Furthermore, while computing a (discrete) plan, some geometric
constraints on continuous trajectories can be embedded in domain description, to en-
sure feasible (collision-free) plans.
REACT! utilizes two sorts of automated reasoners: SAT solvers (e.g., CHAFF [1],
MINISAT [2], MANYSAT [3]) and ASP solvers (e.g., CLASP [4]). According to SAT [5],
the idea is to formalize (in general NP-hard) computational problems as a set of formu-
las in propositional logic so that the models of this set of formulas characterize the
solutions of the given problem; and compute the models using SAT solvers. According
to ASP [6,7], the idea is similar; though the problems are represented in a different log-
ical formalism where the formulas (called rules) look different, have a non-monotonic
meaning and the models (called answer sets [8]) are computed using ASP solvers. Both
SAT and ASP have been applied in various real-world applications. For instance, SAT
has been used for software and hardware veriﬁcation [9] and planning [10]; ASP has
been applied to biomedical informatics [11], space shuttle control [12], workforce man-
agement [13], etc.
Although both SAT and ASP provide efﬁcient solvers and their formalisms are gen-
eral enough to represent various kinds of computational problems, their formalisms do
not provide special structures for a systematic formalization of dynamic domains. To
facilitate the use of such general knowledge representation formalisms for representing
dynamic domains, some other form of logic-based formalisms, called action descrip-
tion languages [14], have been introduced. Further, to be able to use SAT/ASP solvers
for reasoning about actions and change, sound transformations from action description
languages to SAT and ASP have been developed [15,16,17].
REACT! provides an interactive interface to systematically represent actions and
change in the action language C+ [17]. It also allows the users to solve planning prob-
lems using state-of-the-art SAT/ASP solvers. In that sense, REACT! helps robotic re-
searchers to learn, understand, and use high-level representation and reasoning con-
cepts, formalisms, methods, and reasoners.
2
Reasoning about Actions and Change
For an agent to act intelligently in a dynamic domain, one of the essential high level
cognitive functions for that agent is reasoning about its actions and the change that
is directly or indirectly caused by these actions. For instance, AI planning is one of
such reasoning tasks: a robotic agent should be able to autonomously ﬁnd a sequence
of actions to execute, in order to reach a given goal from the initial state she is at. To
perform reasoning tasks, an agent should know about which actions she can execute, as
well as how these actions change the world. For that, we can describe the actions of the
agent in a logic-based formalism so that the agent can autonomously perform reasoning
tasks (e.g., ﬁnd plans) by using an automated reasoner based on logic-based inferences.
On the other hand, representing actions of an agent in a logic-based formalism re-
quires some background in logic as well as the speciﬁc representation language. Con-
sider, for instance, a mobile robot that can go from one location to another location, as
well as pick and place some boxes in these locations. States of the world can be de-
scribed by means of three ﬂuents (i.e., predicates whose truth value may change over
time): one describing the location of the robot, one describing the locations of objects,
another describing whether the robot is holding an object or not. Then, we describe the
action of a robot going to a location y, by representing the preconditions (i.e., the robot
is not at y) and the direct effects (i.e., the robot is at y). We also describe the indirect
effects of actions, and state/transition constraints, like:
Ramiﬁcations: If the robot is holding a box b, and the robot goes to some location
y, then as an indirect effect of this action the location of the box b becomes y as well.
State Constraints: Every robot (or box) cannot be at two different locations, and
two boxes cannot be at the same location at any state of the world.
Further, we need the commonsense law of inertia: if an action does not change the
location of the robot (resp. a box), then the robot’s (resp. a box’s) location remains to
be the same.
Figures 1, 2, and 3 show parts of the robot’s domain, in particular the representation
of the action of going to a location, in SAT (by a set of clauses), in ASP (by a set of
rules) and in C+ (by a set of causal laws), respectively. In these representations, y ranges
over locations {L1,...,Lm}, and b ranges over boxes {B1,...,Bn}. As you can see from
these formulations, it is hard to understand which clauses in the SAT formulation de-
scribe preconditions and which clauses represent direct effects. The ASP formulation is
more concise and it is slightly easier to understand each rule; however, it is still hard to
ﬁgure out which kind of rules (one with nothing on the left-hand-side of the arrow, or
one with some atom) to use for representing what. The C+ formulation is closer to nat-
ural language and easier to understand: causal laws of the form nonexecutable a if f
describe preconditions ¬ f of an action a; causal laws of the form a causes f describe
direct effects f of an action a; and causal laws of the form caused f if g describe rami-
ﬁcations of actions. However, we still need to know about the syntax and semantics of
formulas in C+ to formalize a robotic domain.
Moreover, once the domain is represented and the reasoning problem is speciﬁed
formally, to solve a given planning problem, we need to know the speciﬁc usage and/or
The robot cannot be at two different locations:
¬atRobo(x,t)∨¬atRobo(y,t)
(x < y)
An object b cannot be at two different locations:
¬atObj(b,x,t)∨¬atObj(b,y,t)
(x < y)
Preconditions of goto(y,t):
¬goto(y,t)∨¬atRobo(y,t)
¬goto(y,t)∨¬atObj(b,y,t)
Direct effects of goto(y,t):
¬atRobo(y,t +1)∨goto(y,t)∨atRobo(y,t +1)
¬atRobo(y,t +1)∨goto(y,t)∨atRobo(y,t)
atRobo(y,t +1)∨¬goto(y,t)
atRobo(y,t +1)∨¬atRobo(y,t +1)∨¬atRobo(y,t)
Ramiﬁcations:
¬atObj(b,y,t)∨holding(b,t)∨atObj(b,y,t)
¬atObj(b,y,t)∨holding(b,t)∨atObj(b,y,t −1)
¬atObj(b,y,t)∨atRobo(y,t)∨atObj(b,y,t)
¬atObj(b,y,t)∨atRobo(y,t)∨atObj(b,y,t −1)
atObj(b,y,t)∨¬holding(b,t)∨¬atObj(b,y,t)
atObj(b,y,t)∨¬atObj(b,y,t)∨¬atObj(b,y,t −1)
...
Fig. 1. Describing the robot’s domain in SAT
The robot cannot be at two different locations:
←atRobo(x,t),atRobo(y,t)
(x < y)
An object b cannot be at two different locations:
←atObj(b,x,t),atObj(b,y,t)
(x < y)
Preconditions of goto(y,t):
←goto(y,t),atRobo(y,t)
←goto(y,t),atObj(b,y,t)
Direct effects of goto(y,t):
atRobo(y,t +1) ←goto(y,t)
Ramiﬁcations:
atObj(b,y,t) ←holding(b,t),atRobo(y,t)
...
Fig. 2. Describing the robot’s domain in ASP
command lines of the relevant automated reasoners. Therefore, it is challenging and
time-consuming to learn how dynamic domains can be represented in such a logic-
based formalism, and how automated reasoners can be used to solve planning problems
(and other reasoning problems, such as prediction and postdiction).
We would like to enable the robotic researchers to start using automated reasoners
to solve various planning problems in dynamic domains with robotic agents, and assist
Preconditions of goto(y):
nonexecutable goto(y) if atRobo = y
nonexecutable goto(y) if atObj(o) = y
Direct effects of goto(y):
goto(y) causes atRobo = y
Ramiﬁcations:
caused atObj(b) = y if holding(b)∧atRobo = y
...
Fig. 3. Describing the robot’s domain in C+
Fig. 4. A precondition of goto(L1) (i.e., the robot is not already at L1), and a direct effect of
goto(L1) (i.e., the robot is at L1).
them to have a better understanding of the concepts on reasoning about actions and
change, so that they can build/use robots that are furnished with deliberate reasoning
capabilities to autonomously perform some tasks. With this motivation, we have built
an interactive tool that guides robotic researchers
– to represent dynamic domains in a generic way (so they do not have to know about
a speciﬁc action description language),
– to embed continuous geometric reasoning in discrete task planning in a modular
way (so they do not have to modify the source codes of relevant planners or imple-
ment new hybrid planning algorithms), and
– to solve planning problems using various planners/reasoners (so they do not have
to know any speciﬁcs of these systems).
3
REACT!
REACT! is an interactive tool that helps the users represent a dynamic domain in a
logic-based formalism, and solve a planning problem in this domain using an automated
Fig. 5. A planning problem.
reasoner. It guides the users with an interactive user-interface providing explanations
and examples, so that the users do not have to know about the underlying formalism
or the reasoner. REACT! also assists users to have an understanding of fundamental
concepts in knowledge representation and reasoning.
For instance, the preconditions and direct effects of the action of going to a location
are described in REACT! as depicted in Figure 4. As seen in the upper inset, the user
describes (in the left part of the user-interface) the following precondition of goto(L1):
the robot is not already at L1. While describing the precondition, the user can use the
variables and the object constants (shown on the right part of the user-interface) de-
clared earlier, with the aid of auto-completion. After the user adds this precondition, it
is displayed in the syntax of C+ on the right part of the interface, as shown in the lower
inset in Figure 4. The question mark symbols on the user-interface provide information
about what preconditions and effects are, how they look like, with examples.
Once the action domain is described, the user checks whether the domain descrip-
tion is consistent. If the description is consistent, then the user continues with the spec-
iﬁcation of a planning problem by means of an initial state and a goal state, as shown in
the upper inset in Figure 5. This problem can then be solved by an automated reasoner.
Currently, the underlying formalism of REACT! is the action description language
C+. The user can choose one of the state-of-the-art SAT solvers among CHAFF, MIN-
ISAT and MANYSAT, or the state-of-the-art ASP solver CLASP as an automated rea-
soner. If the user chooses a SAT solver, REACT! uses CCALC [18] to transform the
action domain description in C+ to a set of clauses, and calls the SAT solver to ﬁnd a
plan for the given problem with respect to the described action domain, as shown in the
lower inset in Figure 5. If the user chooses an ASP solver, REACT! uses the program
CPLUS2ASP [19] to transform the action domain description in C+ to a set of rules in
ASP, and calls CLASP to ﬁnd a plan for the given problem with respect to the described
action domain (Figure 5).
Fig. 6. Embedding geometric reasoning (i.e., to check existence of a collision-free path) as part
of a precondition of goto(L1): if the robot is at location L2 then it can goto (L1) if there is a
collision-free path between L1 and L2.
3.1
Hybrid Planning with REACT!
REACT! allows integration of continuous geometric constraints (e.g., by calling a mo-
tion planner to check whether going to some location by means of a continuous trajec-
tory is feasible without colliding to any static object) as well as temporal constraints
(e.g., by deﬁning durative actions and imposing deadlines for completion of tasks) into
discrete high-level representation of a domain description, so that the discrete plan com-
puted by REACT! is feasible. Such an integration is possible by means of “external
predicates” [20,21]—predicates that are not part of the action domain description (i.e.,
different from ﬂuents and actions) but that are deﬁned externally in some program-
ming language (e.g., C++, Prolog). Integrating geometric reasoning (in preconditions
of actions) and temporal reasoning (in planning problems) are possible thanks to the ex-
pressive input language of CCALC. Essentially, CCALC performs some preprocessing
of external predicates, while transforming causal laws into propositional logic formulas.
The ASP solver CLASP also supports external predicates.
The lower inset in Figure 6 shows an example of integrating collision checking
as part of the preconditions of the action of going to some location L1 (from a location
L2), by means of an external predicate pathExists(L1,L2) declared earlier as in the upper
inset. By this way, geometric reasoning is embedded in high-level representation: while
computing a task-plan, the reasoner takes into account geometric models and kinematic
relations by means of external predicates implemented for geometric reasoning. In that
sense, the geometric reasoner guides the reasoner to ﬁnd feasible solutions.
An interesting example of hybrid planning with temporal constraints is given for the
housekeeping domain in [22]. In this domain, not only an external predicate pathExists
is used to check geometric feasibility of action of going to some location, but also a
second external predicate timeEstimate is utilized to estimate the time it will take for
the robot to traverse the path. Then, while describing the planning problem, temporal
Fig. 7. REACT! provides dynamic simulations for solutions of Tower of Hanoi problem instances.
constraints, for instance the constraint that the total time required to complete the plan
should be less than a predeﬁned value, can be added to the speciﬁcation of a planning
problem (the upper inset of Figure 5).
Hybrid planning with geometric and temporal constraints has been applied in var-
ious domains (e.g., cognitive factories [23], service robotics [22,24], robotic manipu-
lation [25,26]) in the spirit of cognitive robotics [27]. REACT! allows formulations of
these domains.
3.2
Sample Domains and Simulation Interface of REACT!
To facilitate the use of REACT! for robotic applications and help robotic researchers to
gain experience on fundamental concepts in reasoning about actions and change, RE-
ACT! provides a set of example domains, including Tower of Hanoi and Multi-Agent
Path Planning problems. In each example domain, REACT! provides explanations at
each tab (e.g., about the concepts of a ﬂuent, an action, preconditions/effects of an ac-
tion, ramiﬁcations, static/transition constraints, planning problem), so that the user can
have a better understanding of systematically representing a dynamic domain. Once a
plan is computed by an automated reasoner, REACT! also provides a dynamic simu-
lation for an execution of the plan, using OPENRAVE [28]. For the Tower of Hanoi
example, REACT! also provides a wrapper to execute the plans on a physical KuKa
youBot manipulator through Robot Operating System (ROS). For instance, Figure 7
shows a snapshot of a simulation of a plan computed by MINISAT for a Tower of
Hanoi problem, while a movie of its physical implementation can be viewed from the
following link: http://cogrobo.sabanciuniv.edu/?p=690.
4
Conclusions
We have introduced REACT! as an interactive tool for cognitive robotic applications.
Signiﬁcantly reducing the learning time, REACT! lets robotic researchers to concentrate
on robotic applications, by enabling them to describe action domains systematically and
to solve complex problems relevant to robotics applications using automated reasoners.
REACT! not only assists its users to have a better understanding of the concepts on
reasoning about actions and change, but also provides sample formalizations of some
action domains, as well as dynamic simulations of plans computed by a selected auto-
mated reasoner.
Acknowledgments
This work is partially supported by TUBITAK 111E116.
References
1. Moskewicz, M.W., Madigan, C.F., Zhao, Y., Zhang, L., Malik, S.: Chaff: Engineering an
efﬁcient SAT solver. In: Proc. 38th Design Automation Conf. (DAC). (2001) 530–535
2. E´en, N., S¨orensson, N.: An extensible SAT-solver. In: Proc. 6th Int. Conf. Theory and
Applications of Satisﬁability Testing (SAT). (2003) 502–518
3. Hamadi, Y., Jabbour, S., Sais, L.: ManySAT: A parallel SAT solver. Journal on Satisﬁability,
Boolean Modeling and Computation 6(4) (2009) 245–262
4. Gebser, M., Kaufmann, B., Neumann, A., Schaub, T.: clasp: A conﬂict-driven answer set
solver. In: Proc. of LPNMR. (2007) 260–265
5. Gomes, C.P., Kautz, H., Sabharwal, A., Selman, B.: Cognitive robotics. In: Handbook of
Knowledge Representation, Elsevier (2008)
6. Lifschitz, V.: What is answer set programming? In: Proc. of AAAI, MIT Press (2008)
1594–1597
7. Brewka, G., Eiter, T., Truszczynski, M.: Answer set programming at a glance. Commun.
ACM 54(12) (2011) 92–103
8. Gelfond, M., Lifschitz, V.: Classical negation in logic programs and disjunctive databases.
New Generation Computing 9 (1991) 365–385
9. Velev, M.N., Bryant, R.E.: Effective use of boolean satisﬁability procedures in the formal
veriﬁcation of superscalar and vliw microprocessors. J. Symb. Comput. 35(2) (2003) 73–106
10. Kautz, H.A., Selman, B.: Planning as satisﬁability. In: Proc. of ECAI. (1992) 359–363
11. Erdem, E., Erdem, Y., Erdogan, H., Oztok, U.: Finding answers and generating explanations
for complex biomedical queries. In: Proc. of AAAI. (2011)
12. Nogueira, M., Balduccini, M., Gelfond, M., Watson, R., Barry, M.: An a-prolog decision
support system for the space shuttle. In: Proc. of PADL, Springer (2001) 169–183
13. Ricca, F., Grasso, G., Alviano, M., Manna, M., Lio, V., Iiritano, S., Leone, N.: Team-building
with answer set programming in the Gioia-Tauro seaport. Theory and Practice of Logic
Programming 12 (2012)
14. Gelfond, M., Lifschitz, V.: Action languages. Electronic Transactions on Artiﬁcial Intelli-
gence 2 (1998) 193–210
15. McCain, N., Turner, H.: Satisﬁability planning with causal theories. In: Proc. of KR. (1998)
212–223
16. Lifschitz, V., Turner, H.: Representing transition systems by logic programs. In: Proc. of
LPNMR. (1999) 92–106
17. Giunchiglia, E., Lee, J., Lifschitz, V., McCain, N., Turner, H.: Nonmonotonic causal theories.
Artiﬁcial Intelligence 153(1–2) (2004) 49–104
18. McCain, N., Turner, H.: Causal theories of action and change. In: Proc. 14th Nat. Conf.
Artiﬁcial Intelligence. (1997) 460–465
19. Casolary, M., Lee, J.:
Representing the language of the causal calculator in answer set
programming. In: Proc. of ICLP (Technical Communications). (2011) 51–61
20. McCain, N.: Causality in Commonsense Reasoning about Actions. PhD thesis, UT Austin
(1997)
21. Eiter, T., Ianni, G., Schindlauer, R., Tompits, H.: Effective integration of declarative rules
with external evaluations for semantic-web reasoning. In: Proc. of ESWC. (2006) 273–287
22. Aker, E., Patoglu, V., Erdem, E.: Answer set programming for reasoning with semantic
knowledge in collaborative housekeeping robotics. In: Proc. Int. IFAC Symp. Robot Control
(SYROCO). (2012)
23. Erdem, E., Haspalamutgil, K., Patoglu, V., Uras, T.: Causality-based planning and diagnostic
reasoning for cognitive factories. In: Proc. 17th IEEE Int. Conf. Emerging Technologies and
Factory Automation (ETFA). (2012)
24. Aker, E., Erdogan, A., Erdem, E., Patoglu, V.: Causal reasoning for planning and coordi-
nation of multiple housekeeping robots. In: Proc. 12th Int. Conf. Logic Programming and
Nonmonotonic Reasoning (LPNMR). (2011)
25. Erdem, E., Haspalamutgil, K., Palaz, C., Patoglu, V., Uras, T.: Combining high-level causal
reasoning with low-level geometric reasoning and motion planning for robotic manipulation.
In: Proc. of IEEE Int. Conf. Robotics and Automation (ICRA). (2011)
26. Havur, G., Haspalamutgil, K., Palaz, C., Erdem, E., Patoglu, V.: A case study on the Tower
of Hanoi Challenge: Representation, reasoning and execution. In: Proc. of IEEE Int. Conf.
Robotics and Automation (ICRA). (2013)
27. Erdem, E., Patoglu, V.: Applications of action languages in cognitive robotics. In: Correct
Reasoning. (2012) 229–246
28. Diankov, R.:
Automated Construction of Robotic Manipulation Programs.
PhD thesis,
Carnegie Mellon University, Robotics Institute (August 2010)