Search

Minimally invasive robotic surgery copes with some disadvantages for the surgeon of minimally invasive surgery while preserving the advantages for the patient. Most commercially available robotic systems are telemanipulated with haptic input devices. The exploitation of the haptics channel, e.g., by means of Virtual Fixtures, would allow for an individualized enhancement of surgical performance with contextual assistance. However, it remains an open field of research as it is non-trivial to estimate the task context itself during a surgery. In contrast, surgical training allows to abstract away from a real operation and thus makes it possible to model the task accurately. The presented approach exploits this fact to parameterize Virtual Fixtures during surgical training, proposing a Shared Control Parametrization Engine that retrieves procedural context information from a Digital Twin. This approach accelerates a proficient use of the robotic system for novice surgeons by augmenting the surgeon’s performance through haptic assistance. With this our aim is to reduce the required skill level and cognitive load of a surgeon performing minimally invasive robotic surgery. A pilot study is performed on the DLR MiroSurge system to evaluate the presented approach. The participants are tasked with two benchmark scenarios of surgical training. The execution of the benchmark scenarios requires basic skills as pick, place and path following. The evaluation of the pilot study shows the promising trend that novel users profit from the haptic augmentation during training of certain tasks.

Intuitive robot programming for non-experts will be essential to increasing automation in small and medium-sized enterprises (SMEs). Programming by Demonstration (PbD) is a fast and intuitive approach, whereas programs created with Task-Level Programming (TLP) are easy to understand and flexible in their execution. In this paper, we propose an approach which combines these complementary advantages of PbD and TLP. Users define complete task-level programs including all parameters through PbD alone. Therefore, we call this approach Task-Level Programming by Demonstration (TLPbD). TLPbD extends skill-based approaches by enabling experts to semantically annotate robot skills with their conditions and effects, which facilitates online skill recognition from pure demonstrations by a non-expert. In a user study with 21 participants, the approach is compared with an existing intuitive TLP approach. The results show that the new approach drastically reduces the programming time while at the same time being more intuitive, reducing mental load, and achieving the same or even better skill sequences.

We propose a fault-tolerant estimation technique for the six-DoF pose of a tendon-driven continuum mechanisms using machine learning. In contrast to previous estimation techniques, no deformation model is required, and the pose prediction is rather performed with polynomial regression. As only a few datapoints are required for the regression, several estimators are trained with structured occlusions of the available sensor information, and clustered into ensembles based on the available sensors. By computing the variance of one ensemble, the uncertainty in the prediction is monitored and, if the variance is above a threshold, sensor loss is detected and handled. Experiments on the humanoid neck of the DLR robot DAVID, demonstrate that the accuracy of the predicted pose is significantly improved, and a reliable prediction can still be performed using only 3 out of 8 sensors.

As robots get closer to human environments, a fundamental task for the community is to design system behaviors that foster trust. In this context, we have posed the "Green Button Challenge": every robot should have a green button that, when pressed, makes the robot explain what it is doing and why, in natural language. In this paper, we motivate why explainability is important in robotics, an why explicit knowledge representations are essential to achieving it. We highlight this with a concrete proof-of-concept implementation on our humanoid space assistant Rollin' Justin, which interprets its PDDL plans to explain what it is doing and why.

Reinforcement learning (RL) enables robots to learn skills from interactions with the real world. In practice, the unstructured step-based exploration used in Deep RL -- often very successful in simulation -- leads to jerky motion patterns on real robots. Consequences of the resulting shaky behavior are poor exploration, or even damage to the robot. We address these issues by adapting state-dependent exploration (SDE) to current Deep RL algorithms. To enable this adaptation, we propose three extensions to the original SDE, which leads to a new exploration method generalized state-dependent exploration (gSDE). We evaluate gSDE both in simulation, on PyBullet continuous control tasks, and directly on a tendon-driven elastic robot. gSDE yields competitive results in simulation but outperforms the unstructured exploration on the real robot.

This paper gives an overview of the control and computing architectures of current lightweight robots and their demands on on board data processing units. On Earth, robots are an established technology in our daily life. They are getting more and more aware of their surroundings, which allows their use in many new domains. Collaborative workplaces where a robotic co-worker directly interacts with a human are already state of the art. Current research on machine learning and Artificial Intelligence (AI) in the robotic domain will introduce many new applications for robotic systems. These developments also have a great potential to be used in future space missions, for example on-orbit servicing of satellites or maintenance of space stations. However, this induces certain requirements on involved computing hardware that future on-board data processing units will also have to deal with.

International audience; In the near future, robots collaborating with human operators in industries will need more and more anticipation capabilities to properly react to human actions and provide efficient collaboration. To achieve this goal, new technologies are needed that not only estimate the motion of the humans, but that fully describe the whole-body dynamics of the interaction and that can predict its outcome. These hardware and software technologies are the goal of the European project AnDy. The AnDy project leverages existing technologies to endow robots with the ability to control physical collaboration through intentional interaction in order to maximize ergonomics for the user. To achieve this goal, AnDy relies on three technological and scientific breakthroughs. First, AnDy innovates the way of measuring human whole-body motions by developing a wearable AnDySuit, which tracks motions and records forces. Second, AnDy develops the AnDyModel, which combines ergonomic models with cognitive predictive models of human dynamic behavior in collaborative tasks, learned from data acquired with the AnDySuit. Third, AnDy proposes AnDyControl, an innovative technology for assisting humans through predictive physical control based on AnDyModel. By measuring and modeling human whole-body dynamics, AnDy will provide robots with a new level of awareness about human intentions and ergonomics. By incorporating this awareness on-line in the robot's controllers, AnDy paves the way for novel applications of physical human-robot collaboration in manufacturing, health-care, and assisted living. We present the goals and methods of the AnDy project, as well as first year results. Technical advances on the AndySuit – using inertial motion capture and sensorized shoes – along with the development of an on-line inverse dynamics software tool now allows real-time monitoring of human dynamics. The information provided thereby is used in the controller of a robot physically interacting with the human, so that the robot reactively adapts its movement to the human’s movement. Experiments with a real robot have shown promising results. The next step is to couple the robot controller with an automatic ergonomic assessment tool. Thereby, the robot will be able to detect and anticipate critical situations, and will react in order to optimize the ergonomics of the human movement.

International audience; Future robots will need more and more anticipation capabilities, to properly react to human actions and provide efficient collaboration. To achieve this goal, we need new technologies that not only estimate the motion of the humans, but that fully describe the whole-body dynamics of the interaction and that can also predict its outcome. These hardware and software technologies are the goal of the European project AnDy. In this paper, we describe the roadmap of AnDy, which leverages existing technologies to endow robots with the ability to control physical collaboration through intentional interaction. To achieve this goal, AnDy relies on three technological and scientific breakthroughs. First, AnDy will innovate the way of measuring human whole-body motions by developing the wearable AnDySuit, which tracks motions and records forces. Second, AnDy will develop the AnDyModel, which combines ergonomic models with cognitive predictive models of human dynamic behavior in collaborative tasks, learned from data acquired with the AnDySuit. Third, AnDy will propose AnDyControl, an innovative technology for assisting humans through pre-dictive physical control, based on AnDyModel. By measuring and modeling human whole-body dynamics, AnDy will provide robots with a new level of awareness about human intentions and ergonomy. By incorporating this awareness on-line in the robot's controllers, AnDy paves the way for novel applications of physical human-robot collaboration in manufacturing, health-care, and assisted living.

International audience; In a variety of problem domains, such as math and motion planning, humans use a dual strategy of generation and recall to find solutions. 'Generation' uses production rules and models to search for novel solutions to novel problems, whereas 'recall' reuses previously found solutions for similar previously encountered problems. As we expect the advantages of this dual strategy to carry over to the robotics domain, we compare and evaluate generation and recall strategies for motion planning on a set of reaching tasks. The specific implementations we use are the lazy variant of the Rapidly-exploring Random Trees and Dynamic Movement Primitives, and we compare these two methods on the commercially available REEM robot. Quantifying the differences and advantages of these methods constitutes is required to make informed decisions about which approach is most suitable for which application domain and task contexts.