Dialog is a useful way for a robotic agent performing a task to communicate with a human collaborator, as it is a rich source of information for both the agent and the human. Such task-oriented dialog provides a medium for commanding, informing, teaching, and correcting a robot. Robotic agents engaging in dialog must be able to interpret a wide variety of sentences and supplement the dialog with information from its context, history, learned knowledge, and from non-linguistic interactions. We have identified a set of nine system-level requirements for such agents that help them support more effective, efficient, and general task-oriented dialog. This set is inspired by our research in Interactive Task Learning with a robotic agent named Rosie. This paper defines each requirement and gives examples of work we have done that illustrates them.

Recent progress in machine learning has lead to great advancements in robot intelligence and human-robot interaction (HRI). It is reported that robots can deeply understand visual scene information and describe the scenes in natural language using object recognition and natural language processing methods. Image-based question and answering (Q&A) systems can be used for enhancing HRI. However, despite these successful results, several key issues still remain to be discussed and improved. In particular, it is essential for an agent to act in a dynamic, uncertain, and asynchronous envi-ronment for achieving human-level robot intelligence. In this paper, we propose a prototype system for a video Q&A robot “Pororobot”. The system uses the state-of-the-art machine learning methods such as a deep concept hierarchy model. In our scenario, a robot and a child plays a video Q&A game together under real world environments. Here we demonstrate preliminary results of the proposed system and discuss some directions as future works.

Autonomous agents require interfaces to define their interactions with humans. The coupling between agents and humans is often limited, with disjoint goals between the agent interface and its associated autonomous components. This leads to a gap in human interaction relative to agent capabilities. We seek to aid interface designs by clarifying agent capabilities within an interface context. A taxonomy was developed that can help elucidate the agent’s affordances and constraints that guide interface design. Moreover, the descriptors employed in the taxonomy can serve as a common language to support dialog between agent and interface developers, resulting in improved autonomous systems that support human-autonomy coordination.

Motivation impacts people’s lives in a powerful way and is at the heart of a plethora of day-to-day activities and achievement settings, from success at the workplace to learning and acquiring knowledge to trying to quit bad habits. The current work aims to develop an adaptive robot companion that models a user’s daily motivational state and chooses appropriate motivational strategies to keep the user on track for achieving a daily goal. The two main components we are focusing on in this context are creating an ontology-based user model of the person’s motivational states and using an appropriate strategy selection algorithm that chooses the best motivational strategies for the user each day based on the user model’s output. Specifically, we are focusing on the important application domain of physical activity and aim to help early adolescents achieve daily-recommended levels of physical activity. Our human-robot interaction system uses information acquired from the user to feed the user model and physical activity data from a wristband device to inform the strategy selection algorithm.

Advanced robotic technology is opening up the possibility of integrating robots into the human workspace to improve productivity and decrease the strain of repetitive, arduous physical tasks currently performed by human workers. However, coordinating these teams is a challenging problem. We must understand how decision-making authority over scheduling decisions should be shared between team members and how the preferences of the team members should be included. We report the results of a human-subject experiment investigating how a robotic teammate should best incorporate the preferences of human teammates into the team's schedule. We find that humans would rather work with a robotic teammate that accounts for their preferences, but this desire might be mitigated if their preferences come at the expense of team efficiency.

The theory of mind is an important human capability that allows us to understand and predict the goals, intents, and beliefs of other individuals. We present an approach to designing intelligent communicative agents based on modeling theories of mind. This can be tricky because other agents may also have their own theories of mind of the first agent, meaning that these mental models are naturally nested in layers. So, to look for intuitive communicative acts, we recursively apply a planning algorithm in each of these nested layers, looking for possible plans of action as well as their hypothetical consequences, which include the reactions of other agents; we propose that truly intelligent communicative acts are the ones which produce a state of maximum decision theoretic utility according to the entire theory of mind. We implement these ideas using Java and OpenCyc in an attempt to create an assistive AI we call MARTHA. We demonstrate MARTHA's capabilities with two motivating examples: helping the user buy a sandwich and helping the user search for an activity. We see that, in addition to being a personal assistant, MARTHA can be extended to other assistive fields, such as finance, research, and government.

We consider ways to improve the performance of unsupervised plan and activity recognition techniques by considering temporal and object relations in addition to postural data. Temporal relationships can help recognize activities with cyclic structure and are often implicit because plans have degrees of ordering actions. Relations with objects can help disambiguate observed activities that otherwise share a user's posture and position. We develop and investigate graphical models that extend the popular latent Dirichlet allocation approach with temporal and object relations, examine the relative performance and runtime trade-offs using a standard dataset, and consider the cost/benefit trade-offs these extensions offer in the context of human-robot and humancomputer interaction.

In order for robots to interact with humans in the world around them, it is important that they are not just aware of the presence of people, but also able to understand what those people are doing. In particular, interaction involves multiple agents which requires some form of coordination, and this cannot be achieved by acting blindly. The field of plan recognition (PR) studies methods for identifying an observed agent’s task or goal given her action sequence. This is often regarded as the inverse of planning which, given a set of goal conditions, aims to derive a sequence of actions that will achieve the goals when performed from a given initial state. Ram´ırez and Geffner (2009; 2010) proposed a simple transformation of PR problems into classical planning problems for which off-the-shelf software is available for quick and efficient implementations. However, there is a reliance on the observed agent’s optimality which makes this PR technique most useful as a post-processing step when some of the final actions are observed. In human-robot interaction (HRI), it is usually too late to interact once the humans are finished performing their tasks. In this paper, we describe ongoing work two extensions to make classical planning-based PR more applicable to the field of HRI. First, we introduce a modification to their algorithm that reduces the optimality bias’s effect so that long-term goals may be recognized at earlier observations. This is then followed by methods for extracting information from these predictions so that the observing agent may run a second pass of the planner to determine its own actions to perform for a fully interactive system.

In this position paper, we describe the RoboHelper project, its findings and our vision for its future. The long-term goal of RoboHelper is to develop assistive robots for the elderly. The main thesis of our work is that such robots must crucially be able to participate in multimodal dialogues. Contributions of our work to date include the ELDERLY-AT-HOME corpus that we collected and annotated. It consists of 20 task-oriented human-human dialogues between a helper and an elderly person in a fully functional apartment. The unique feature of the corpus is that in addition to video and audio, it includes recordings of physical interaction. Based on this data, we have demonstrated the crucial role that Haptic-Ostensive (H-O) actions play in interpreting language and uncovering a person's intentions. H-O actions manipulate objects, but they also often perform a referring function. Our models were derived on the basis of manually annotated categories. Additional experiments show that we can identify H-O actions using the physical interaction data measured through an unobtrusive sensory glove developed as part of the project. In future work, we will derive models for the robot to decide what to do next (as opposed to interpreting what the interlocutor did); explore other types of physical interactions; and refine preliminary implementations of our models on the Nao robotic platform.

Our work is aimed at service robots deployed in human environments that will need many specialized object manipulation skill. We believe robots should leverage end-users to quickly and efficiently learn the affordances of objects in their environment. Prior work has shown that this approach is promising because people naturally focus on showing salient rare aspects ofthe objects (Thomaz and Cakmak 2009). We replicate these prior results and build on them to create a semi-supervised combination of self and guided learning.We compare three conditions: (1) learning through self-exploration, (2) learning from demonstrations providedby 10 naive users, and (3) self-exploration seeded with the user demonstrations. Initial results suggests benefits of a mixed initiative approach.