A cognitive architecture for social robots to exchange knowledge about the world and the self.

By 05/02/2020 Blog

Name: Anna Mura, Vicky Vouloutsi

A cognitive architecture for social robots to exchange knowledge about the world and the self.

Society more than ever needs robots performing side by side with humans in different environments and levels of operation, especially in industry and for environmental sustainability.

In our collective imagination, we are prepared to accept “social” robots that can help us to supply those societal/industrial needs that are no longer undertaken by humans. However, we are still far from having cognizant (mindful) robots that understand our needs, and we can trust. Remarkedly, very little is done to transfer principles of perception of the self and the others to autonomous robots that are intended to work side by side with humans. Why is that?

The fundamental problem scientists and engineers struggle with is how much of our understanding of how the mind operates must be transferred to the machine, and how? [Verschure, 2016; Lallee, 2015]. According to the literature, there are two approaches grounded on cognitive architecture that focus on the making of socially interactive robots [T. Fong 2003]: “Functionally- driven robots” where design and functionality are based more on the robot’s “social intelligence appearance” rather than a science-based design. These robots may not require a deep understanding of how the mind operates to build competent robots, i.e., assistive anthropomorphic robots [J. Pineau 2003]; Biologically-grounded robots that are based on theories of natural and social sciences and thus more connected to humans, as they may function using similar principles of perception, decision making, and empathy.

Nevertheless, and in spite of great advances made in the last decade towards solving the worker–robot interaction endeavor in HRI, very few robots work in industry side by side with humans in a collaborative/social manner. And it is clear that developing a cognitive/social architecture to guides our interactions with robots is more critical than previously thought.

The HR-Recycler Project, funded by the H2020 Program of the EU under GA 820742 addresses this challenge by developing a hybrid collaboration environment, where humans and robots will share and undertake at the same time different processing and manipulation tasks, targeting the industrial application of Waste Electrical and Electronic Equipment (WEEE) recycling.

HR-Recycler will advance “Social robotics for safe Human-Robot Collaboration” by capitalizing on the Distributed Adaptive Control (DAC) cognitive architecture [Verschure 2012]. DAC includes a motivation system for social interaction that drives communication and is based on self-regulation and autonomy, and high-level cognitive functions [Lallee 2015, C. Moulin Frier 2017 ].

DAC is organized along four layers (soma, reactive, adaptive and contextual) and three columns (world, self, action). The ‘soma’ designates the body with its sensors, organs and actua- tors. It defines the needs, or self-essential functions (SEF) the organism must satisfy in order to survive. The reactive layer (RL) comprises dedicated behav- iour systems (BS) each implementing predefined sensorimotor mappings serving the SEFs. In order to allow for action selection, task switching and conflict resolution, all BSs are regulated via a, so-called, allostatic controller that sets the internal homeostatic dynamics of BSs relative to overall demands and opportunities [28]. The AL acquires a state space of the agent–environ- ment interaction combining perceptual and behavioural learning constrained by value functions defined by the allostatic control of the RL, minimizing perceptual and behavioural prediction error [29,30]. The contextual layer (CL) further expands the time horizon in which the agent can operate through the use of episodic and sequential short- and long-term memory systems (STM and LTM, respectively). STM acquires conjunctive sensorimotor representations assisted by episodic memory as the agent acts in the world. STM sequences are retained as goal-oriented sequences in LTM when positive value is encountered, as defined by the RL and/or AL. The contribution of stored LTM policies to decision-making depends on four factors: goals, per- ceptual evidence, memory chaining and valence while action selection is further biased by the expected cost of the actions that pertain to reaching a goal state. The content of working memory (WM) is defined by the memory dynamics that represent this four-factor decision-making model. (from P.Verschure 2016)

For collaboration, effective communication is essential for the successful completion of a task. Collaboration is defined as “the mutually beneficial and well-defined relationship of two or more entities to achieve a common goal” [Johal et al. 2014]. Two or more entities that have complementary skills perform common tasks and even share common goals form a team. In Human-Robot Collaboration (HRC) settings, the team is mixed and typically comprises humans and robots working together. For collaboration to be efficient, robots are required to robustly perform a task, be trustworthy, and effectively communicate with the human co-worker. Additionally, we highlight the importance of safe operations, as the robots will be required to function in proximity to humans. For the successful coordination of teamwork, effective communication is critical, as team members may have different mental states [Cohen & Levesque, 2014].

 

“The integration of social cognition and adaptive behaviours to robots in a centralised system will open the door to new HRI and HRC possibilities.”

 

In this context, the SPECS-lab at IBEC as a research group expert in Synthetic Perceptive, Emotive and Cognitive Systems, will contribute to the development of those social and cognitive components needed to build and control “safe Human-Robot Collaboration.” SPECS-lab will do this by expanding DAC’s decision-making model via incorporating into its layered architecture an ethics-based engine responsible for the safe and robust operation of all the components of HR-Recycler.

In addition, the integration of the HR-Recycler human user model together, with social cognition, adaptive behaviours, and learning from human input, will create an environment where the robot will share its current knowledge with the human co-worker. This will help regulate the robot state-space, by either confirming its current knowledge or “adding” into the contextual/cognitive layer of the DAC architecture the appropriate information.

Interestingly the HR-Recycler project will not be using humanoid robots, making the adaptation of the robots’ social and communicative skills to a human co-worker, and vice versa, more challenging but essential for efficient collaboration in industry and manufacturing.

In the field of HRI, we have not found extensive studies to deliver novel ways of interaction suitable for non- anthropomorphic robots. The HR-Recycler project seeks to address the gap in communication for non- anthropomorphic robots based on theories drawn from human communication [Levinson, 2006]. Both Human-Human Interaction (HHI) and Human-Robot Collaboration (HRC) domains require knowledge about the principles and characteristics of social interaction. On the one hand, HHI can benefit from HRI experiments as studies with social robots can act as the testing ground for theoretical explanations. On the other hand, HRC can benefit from studies of HHI, as information acquired from this domain will inform adaptive systems that allow robots to interact with humans fluently. HR-Recycler will advance the field of HRI and HRC by including mechanisms that underlie social competence in a broader range of non-human social behaviors for communication and collaboration.

References:

Leave a Reply