Sensitive Robotics Laboratory
The robots perform manipulation tasks based on the contact information provided by the tactile sensors. The robot Obrero is capable of performing autonomous whole body grasps among other tasks. The robot Gobot is capable of perfoming precision manipulation relaying in tactile feedback. The robot Tactico is capable of working in more cluttered environments. For details refer to the individual links.
Obrero is the first robot using the algoritms of Sensitive Manipulation. These algoritms enabled by the hardware and the tactile sensors. It has 6 DOF compliant arm and a 4 DOF compliant hand. The arm compliance is given by the series elastic actuators (SEAs). The hand has two levels of compliance. One given by the SEAs and one by the tactile sensors.
The next step to test for Sensitive Manipulation was precision manipulation. In this approach tactile feedback allows the robot to manipulate small and slippery objects (GO stones). The position of robot's fingers are not controllable with high precision. Therefore, the task excecution depends on tactile feedback. The approach was succesfully tested.
The succcesor of Obreo and Gobot. This robot was built to handle more complex tasks where the enviromnent is cluttered with objects. The robot arm and and is covered by tactile sensors. The form factor is the one of an industrial arm. More details will be posted soon.
The problem of walking is very similar to manipulation where the contact with the floor provides information for taking the next step while maintaining contact. Most of the approaches to robotic walking focus mainly in balancing minimizing the effect of the feet. In some cases the feet are considered a point of contact. This simplification was used in manipulation as well. In this approach, we use the contact information to develop robust walking algorithms.
The Robot Caminante is the first platform to develop Sensitive Walking. The robot has 6 DOF on each leg. All the joints are driven by SEAs. The feet has an articulated toe and it is covered by tactile sensors. The algoritms take advantage of the contact information to make the robot walk reliably.
Robotics vision has some additional challenges compared to computer vision. In general, the robotic platform and the camera moves and shakes, which makes the images no ideal for computer vision algorithms. In this approach we embrace and promote motion to obtain information.
This robotic head was developed with cameras capable of controlling their zoom, focus, and orientation. In addition, the light intensity is controlled using shades. The main idea behind this approach is taking advantage of the optical DOFs provided by the robotic head.
Flying can be described as manipulation the air. In this approach (Sensitive Flying) as in Sensitive Manipulation we feel the contact with the object to manipulate. Consequently, we expect to be more "dexterous" at flying. We have taken the first steps for this approach building the robotic platform.
This robot flies by feeling the air currents and adapting to the changes. We have implemented our first version. More information will be available after the patent process.
There is a lot of work on how to navigate using GPS or similar position systems. In general robots can navigate in corridors or clear spaces such as roads. However, then the environment is cluttered, the robot usually gets stuck because an expected part of the robot has made contact with an obstacle. For instance, it is very likely that the robot Roomba will get stuck. In this research, address this problem using Sensitve Navigation, where the robot is aware of its body by using tactile feedback.
This robot prototype has its body covered by tactile sensors. It also has whiskers in its front to detect objects that are distant. The robot also has cameras in front.
The features needed to develop useful tactile sensors for robotics are not well understood. In this research we have studied the behavior of compliant tactile sensors that have cruccial characteristics for robotics. These characteristics have been determined after a number of experiments in actual robots performing manipulation.
Compliant Tactile Sensor study
In this research we have investigated the characteristic of a compliant tactile sensor that was designed after experimenting with robots Obrero and GoBot. A Finite Element Model has been created and validated to facilitate the design of tactile sensors. The tactile sensor is patented.
Robots with tactile sensors are difficult to simulate using standard software. There are a number of challenges to implement this simulation. The task is even more complex when the sensors are deformable. In this research we propose a tactile sensor model to implement the simulations.
Compliant tactile sensor: abstraction model
In this research, we present an abstraction model for the sensor that is tested in a simulated manipulation environment. This kind of simulation will enable us to develop algoritms for Sensitive Robotics faster.
The information provided by large arrays of tactile sensors enable the robots to perform complex task. However, we are only starting to explore the richness of this information. New techniques need to be developed to take advantage of these information. These algorithms will impact different areas of robotics that share similar data structures. In this research we have started to explore this area with data available from our own robots.
Tactile data processing
A number of different manchine learning techniques are being used to extract the information available from tactile feedback.