Symbolic planning models allow decision-making agents to sequence actions in arbitrary ways to achieve a variety of goals in dynamic domains. However, they are typically handcrafted and tend to require precise formulations that are not robust to human error. Reinforcement learning (RL) approaches do not require such models, and instead learn domain dynamics by exploring the environment and collecting rewards. However, RL approaches tend to require millions of episodes of experience and often learn policies that are not easily transferable to other tasks. In this paper, we address one aspect of the open problem of integrating these approaches: how can decision-making agents resolve discrepancies in their symbolic planning models while attempting to accomplish goals? We propose an integrated framework named SPOTTER that uses RL to augment and support ("spot") a planning agent by discovering new operators needed by the agent to accomplish goals that are initially unreachable for the agent. SPOTTER outperforms pure-RL approaches while also discovering transferable symbolic knowledge and does not require supervision, successful plan traces or any a priori knowledge about the missing planning operator.

Human-robot communication in situated environments involves a complex interplay between knowledge representations across a wide variety of modalities. Crucially, linguistic information must be associated with representations of objects, locations, people, and goals, which may be represented in very different ways. In previous work, we developed a Consultant Framework that facilitates modality-agnostic access to information distributed across a set of heterogeneously represented knowledge sources. In this work, we draw inspiration from cognitive science to augment these distributed knowledge sources with Short Term Memory Buffers to create an STM-augmented algorithm for referring expression generation. We then discuss the potential performance benefits of this approach and insights from cognitive science that may inform future refinements in the design of our approach.

In this paper we describe moral quasi-dilemmas (MQDs): situations similar to moral dilemmas, but in which an agent is unsure whether exploring the plan space or the world may reveal a course of action that satisfies all moral requirements. We argue that artificial moral agents (AMAs) should be built to handle MQDs (in particular, by exploring the plan space rather than immediately accepting the inevitability of the moral dilemma), and that MQDs may be useful for evaluating AMA architectures.

Artificial agents will need to be aware of human moral and social norms, and able to use them in decision-making. In particular, artificial agents will need a principled approach to managing conflicting norms, which are common in human social interactions. Existing logic-based approaches suffer from normative explosion and are typically designed for deterministic environments; reward-based approaches lack principled ways of determining which normative alternatives exist in a given environment. We propose a hybrid approach, using Linear Temporal Logic (LTL) representations in Markov Decision Processes (MDPs), that manages norm conflicts in a systematic manner while accommodating domain stochasticity. We provide a proof-of-concept implementation in a simulated vacuum cleaning domain.

The Case for Explicit Ethical Agents

The science fiction writer Isaac Asimov was among their designers and make sensitive determinations a handful of visionaries who anticipated the ethical about what should be done (for example, when ethical challenges of deploying autonomous robots in principles are in conflict, they can attempt to human societies. For contexts where (Asimov 1942) were specifically designed to enable informing others of one's intention and reasoning is robots to operate safely in human physical and social crucial, these agents could then also express their reasoning environments, for these laws specify the fundamental in natural language. The key question then is societal obligations any robot has, in order of priority: whether we need such explicit ethical agents or (1) A robot may not injure a human being or, whether current implicit ethical agents are sufficient. Rather, all of the ethical its own existence as long as such protection does not behaviors in such agents will be the result of the conflict with the First or Second Law. Explicit ethical agents, on arise from attempts to apply the laws in different the other hand, will require special representations as well as inference schemes in the cognitive system morally charged situations.

The ability to refer to entities such as objects, locations, and people is an important capability for robots designed to interact with humans. For example, a referring expression (RE) such as “Do you mean the box on the left?” might be used by a robot seeking to disambiguate between objects. In this paper, we present and evaluate algorithms for Referring Expression Generation (REG) in small-scale situated contexts. We first present data regarding how humans generate small-scale spatial referring expressions (REs). We then use this data to define five categories of observed small-scale spatial REs, and use these categories to create an ensemble of REG algorithms. Next, we evaluate REs generated by those algorithms and by humans both subjectively (by having participants rank REs), and objectively, (by assessing task performance when participants use REs) through a set of interrelated crowdsourced experiments. While our machine generated REs were subjectively rated lower than those generated by humans, they objectively significantly outperformed human REs. Finally, we discuss the main contributions of this work: (1) a dataset of images and REs, (2) a categorization of observed small-scale spatial REs, (3) an ensemble of REG algorithms, and (4) a crowdsourcing-based framework for subjectively and objectively evaluating REG.

Machine learning's advances have led to new ideas about the feasibility and importance of machine ethics keeping pace, with increasing emphasis on safety, containment, and alignment. This paper addresses a recent suggestion that inverse reinforcement learning (IRL) could be a means to so-called "value alignment.'' We critically consider how such an approach can engage the social, norm-infused nature of ethical action and outline several features of ethical appraisal that go beyond simple models of behavior, including unavoidably temporal dimensions of norms and counterfactuals.  We propose that a hybrid approach for computational architectures still offers the most promising avenue for machines acting in an ethical fashion.

A major challenge for robots interacting with humans in realistic environments is handling robots' uncertainty with respect to the identities and properties of the people, places, and things found in their environments: a problem compounded when humans refer to these entities using underspecified language. In this paper, we present a framework for generating clarification requests in the face of both pragmatic and referential ambiguity, and show how we are able to handle several stages of this framework by integrating a Dempster-Shafer (DS)-theoretic pragmatic reasoning component with a probabilistic reference resolution component.

One of the hallmarks of humans as social agents is the ability to adjust their language to the norms of the particular situational context. When necessary, they can be terse, direct, and task-oriented, and in other situations they can be more indirect and polite. For future robots to truly earn the label “social,” it is necessary to develop mechanisms to enable robots with NL capabilities to adjust their language in similar ways. In this paper, we highlight the various dimensions involved in this challenge, and discuss how socially-sensitive natural-language generation can be implemented in a cognitive, robotic architecture.