Programming robots is outdated - future robots will be
instructed, not programmed!
We are putting software behind this vision with our integrated
cognitive robotic control architectures that enables
task-based spoken natural language interactions with any kind of
Our software runs on any device with a Java VM and can be
quickly and easily customized to different robots and tasks, using
standard APIs as well as custom integrations, with interfaces to
ROS, JAUS, and other middleware.
It allows users to instruct new tasks and goals through natural
language dialogues and also allows users to teach the robot new
knowledge about objects, activities and task quickly in "one
The demo videos below showcase some of the capabilities of our
system; more demos can be viewed on the Products page.
This video demonstrates task-based dialogues with an
autonomous golf cart (work performed in conjunction with
Indiana University Bloomington). Here, the robot uses
pragmatic inference to determine that the human user has a
false belief about the robot's current goals and asks for
This video demonstrates the system's ability to acquire new
knowledge about objects, object parts, object functions,
actions, and normative action restrictions from single
instructions (work performed in collaboration with the Tufts
Human-Robot Interaction Laboratory). This is different from
statistics-based approaches that require the collection of a
large amount of training data for the robot to learn something
new. No training data is needed for our system!
This video shows a unique capability of our architecture to
share knowledge among multiple robots quickly and efficiently:
if you teach one robot something new, all other robots know it
as well even if they have different physical embodiments (work
performed in collaboration with the Tufts Human-Robot
This video shows our ability to take a fully teleoperated
robot that was never meant for autonomous operation (and does
not provide much of a standardized interface beyond some basic
JAUS teleoperation commands) and "autonomize" it by using the
existing interfaces wherever possible and adding new sensors
when necessary to allow the robot to perform novesl tasks
fully autonomously (work performed in collaboration with
Indiana University Bloomington).
We offer a unique software platform that turns any robot into an
interactive, teachable robot that can be directly instructed by
its users and taught in natural language without any additional
Different licensing models are available based on the required
functionality and level of instruction (e.g., what types of
knowledge users are allowed to teach and whether they are allowed to
change previously learned knowledge).
Assistive robots Our software can be used to
instruct robots a broad set of assistive tasks in various
health and elder care settings. This brief demo video shows
our system learning to perform a simple feeding task where a
caregiver initially positions the arm to allow for food
delivery to the subject. The robot also learns to recognize
and manipulate carrots from instructions.
Our software can be used to instruct a broad set of service
tasks typical for the retail and hospitality industry, as well as many
office settings. This brief demo video shows our system
learning to perform a simple delivery task in a multi-robot
system where a fixed-mounted arm works together with a mobile
platform to jointly perform a delivery task.
We offer complete custom software solutions for robotic tasks on any
robotic platform with standardized interfaces to sensors and
We offer to integrate our software with existing robot platforms
with standardized interfaces (e.g., ROS).
We offer to adapt our software components to specific customer
requirements (e.g., adapting language and corresponding concepts
to specific tasks).
Custom Software Configuration and Setup
We offer to configure, install, and set up our software at the
customer site to ensure everything works as expected.
Software Subscriptions and Updates
We offer different models for obtaining software updates,
through one-time upgrade fees, or automatic upgrades through
We have two decades of research experience on
autonomous robots with natural language capabilities and formal
human subjects evaluations of mixed initiative human-robot
systems, resulting in peer-reviewed publications in top
scientific venues and successful deployment of architectures in
government and university research labs in North America and
We are now excited about bringing our software to the
Matthias Scheutz, Ph.D, Ph.D. is President and CEO of Thinking
He has extensive experience in designing and
implementing integrated cognitive robotic architectures that
can interact with humans in natural language and quickly
acquire new knowledge from natural language instructions.