To support human-robot interaction, designers are taking a page from philosophy and studying how we work together with one another.

The field of human-robot interaction is vast and complex, addressing perception (e.g., tactile or visual), decision-making (e.g., human-aware planning, supervision), and action (e.g., manipulation, navigation). A framework for the design of autonomous robots’ control systems must therefore enable engagement and interaction with human partners.

What is interesting: Findings related to cognitive psychology and philosophy are providing elements converging with what research in human-robot interaction[1] is proposing for the design of interactive robots’ control architecture.

Consider for example a human and a robot working together to tidy up a tabletop (shown in the figure below). The human needs the bottle, out of reach for him, to put it in a bin. What knowledge does the robot need to have about the human for helping him in an effective way? What processes does it need to handle to manage a successful interaction? How are these processes organized? Conversely, what information should humans possess to understand what a robot is doing and how a robot should make this information available to its human partner? (RELATED CONTENT: Learn about the Robonaut assisting NASA's crew aboard the International Space Station.)

What information should humans possess to understand a robot's actions? And how should a robot make this information available to its human partner? (Credit: LAAS-CNRS)

Interaction and Intentions

From a philosophical standpoint, research that defines the interaction between humans has been ongoing for sometime. Philosophers have explained in some detail how people cooperate and interact when working together. This philosophical theory underlying joint action largely relates to intentions, whereby an individual chooses and commits to a course of action in order to achieve a common goal with another individual.

What knowledge does a robot need in order to reach for an object, like a bottle on a table? (Credit: LAAS-CNRS)

Human intentions, as they relate to carrying out joint actions, can be distinguished in three main stages that include:

  • A distal intentions level (D-intentions) in charge of the dynamics of decision making through temporal flexibility and a high level rational guidance and monitoring of actions;
  • A proximal intentions level (P-intentions) that inherits a plan from the previous level and whose role is to anchor this plan in the situation of action — anchoring that needs to be performed at two levels: temporal anchoring and situational anchoring;
  • A motor intentions level (M-intentions), which encodes the fine-grained details of the action (corresponding to what neuroscientists call motor representations), and is responsible for the precision and smoothness of action execution, and which operates at a finer time scale than either D-intentions or P-intentions.

When compared to proposed models for robotic architectures based on action, this framework defining intentions and how they relate to human interactions drew interesting parallels to work being done in the field of human-robot interaction. In fact, what becomes apparent is a clear convergence between a philosophical account of the structure and dynamics of human action and a robot control architecture dedicated to action — and in particular human-robot joint action.

Robotic Architecture for Achieving Human-Robot Joint Action

In relation to philosophers’ findings on human interaction, the robotics community has addressed the problem of robot control architectures, with the objective of building consistent and efficient robot system structures that integrate perception, decision, and action capacities, and provide for both deliberation and reactivity. Directly corresponding to the aforementioned levels of intention, a three-layered robotics architecture has emerged that defines the following:

  • A decision level that can produce a plan to accomplish a task, as well as supervise its execution, while, at the same time, be reactive to events from the next level below. The coexistence of these two features, a time-consuming planning process and a time-bounded reactive supervisory process, raises the key challenge of their interaction and their integration to balance deliberation and reaction at the decision level. In essence, the supervisory component draws upon the planner, which uses knowledge on the environment and human capacities. The planner is said to be “human-aware” because it can arrange a course of actions that takes human capacities into account. This level feeds the next level with the sequence of actions to be executed.
  • An execution control level that controls and coordinates the execution of functions distributed in operational modules, composing the next level, according to the task requirements to achieve the plan. It is at this level that context-based action refinement is performed.
  • A functional or operational level that includes operational modules embedding robot action and perception capacities for data and image processing and motion control loops.

This type of architecture relies on building representations of actions, goals, and plans grounded in human-human interaction.


In working towards architectures dedicated to human-robot interaction, a three-layer division of the robot control architecture seems to be meaningful not only for the robot operating autonomously, but also interactively, similarly to what philosophers have identified for human-human interaction. Of course, there is much more involved when diving deeper into the philosophical aspects related to developing human-robot interaction, as well as determining precisely the various robot capabilities involved in the overall process of humans and robots achieving tasks collaboratively.

Having a clear understanding of human interactions has become a foundation in creating robot control architectures that address key design elements needed for safe and effective human-robot interaction and that are capable of meeting new challenges as robots become more advanced across a range of capabilities, including the ability to learn from increased interaction with humans. Most recently, issues such as ethically aligned design concerns and transparency in robotic interactions so that they are predictable have come to the forefront. These additional topics provide additional, but essential design guidelines for robot architectures.

What do you think? Will we interact with robots in the same ways that we interact with humans? Place your comments in the form below.

About the Author: Raja Chatila is the Executive Chair for The IEEE Global Initiative for Ethical Considerations in Artificial Intelligence and Autonomous Systems. He is also Professor at Pierre and Marie Curie University in Paris and Director of the Institute of Intelligent Systems and Robotics. His research covers several aspects of cognitive robotics, human-robot interaction, and robot learning.


[1] Clodic, A., Pacherie, E. Alami, R. & Chatila, R. (2017). Key Elements in Human-Robot Joint Action. In R. Hakli & J. Seibt (eds.), Sociality and Normativity for Robots: Philosophical Inquiries into Human-Robot Interactions. Springer.