In order to employ service robots in domestic environments, it is necessary to adapt these robots to the human’s living space in the best possible way. Legged walking machines allow the climbing of stairs and stepping over small obstacles without the need of circumnavigating them. Depending on the specific application a different design with two, four, or six legs will be the most appropriate choice. We decided to build a system with humanoid shape based on two legs, because this allows for stable standing and dynamic walking with a small support area. From a sciencific point of view this system allows for fundamental research on dynamic walking and multi-contact interaction in general human environments.
TORO is a bipedal humanoid robot based on the legs of the former DLR-Biped. For research on multi-contact interaction and dynamic whole body motion, an upper body with two six DOF arms and actuated waist was designed based on the torque controlled drive technology of the DLR lightweight arms, similar to the previous design of the DLR-Biped. In order to fit on the legged basis and to reduce load during walking, a redesign of the hip construction was necessary. The first two shoulder joints are assembled by a segment of the DLR lightweight robot arm, while the remaining segments were customized. For interaction with the environment, the arms are equipped with articulated hands. Since TORO will be used mainly for research on control oriented problems related to whole body dynamics, we decided for rather simple but robust prosthetic hands. These hands allow for robust interaction with the environment, but would be limited in terms of dexterous manipulation. For environment perception and egomotion estimation, the system is equipped by an actuated head with stereo cameras, IMU, and Kinect sensor. TORO has a total weight of about 75 kg and a height of about 160 cm.
The main data and distribution of joints in TORO are outlined in the following table...
If biped robots should interact with humans in everyday tasks and environments, they must have a proper control system that allows them to balance (compliantly) in the presence of unknown external perturbations. Such balance is achieved through the application of suitable contact forces to the ground using the finite support area of the feet. Our approach for a posture controller is strongly based on the observation that the problems of grasping an object and balancing a robot are fundamentally similar, in the sense that both try to achieve a desired wrench (on the object in the grasping case, on the robot in the balancing case) based on the application of suitable forces at the contact points (at the fingertips or at the feet). After an unknown perturbation deviates the robot from a desired posture, the controller computes the wrench required to recover the desired position and orientation, according to a compliance control law. This wrench is distributed to predefined supporting contact points at the feet. The forces at these points are computed via a constrained optimization problem that minimizes the contact forces while including friction restrictions and torque limits at each joint.
The same basic ideas are currently being extended to tackle the problem of multi-contact interaction, where the robot can use additional contact points (e.g. hands) to keep the balance or interact with the world, as in the case of crawling through challenging terrain or climbing a ladder.
On rough and uneven terrain (such as stairs, rocks, hills), walking machines are expected to have better locomotion capabilities compared to wheeled machines. Thus, one of our research interests is the development of algorithms for bipedal walking.
Due to the complexity of the full dynamics of a humanoid robot, it is common to use simplified models – so called template models – which represent only parts of the robot’s dynamics (typically the Center of Mass (CoM) dynamics). These template models include Inverted Pendulum (IP => Compass Gait Walking), Spring Loaded Inverted Pendulum (SLIP, typically used for running and bio-inspired walking) and Linear Inverted Pendulum (LIP; used in most state-of-the-art walking robots; assumes a constant CoM height). Our work was traditionally based on the LIP. To handle the robots CoM dynamics, we used the concept of 'Capture Point' (introduced by Jerry Pratt  in 2006). Recently (see thumbnail of paper below), we extended the concept of 'Capture Point' to 3D, such that the CoM height is no longer constrained to be constant and walking over three-dimensional ground surfaces is facilitated (see figure).
The embedding of the concept of Capture Point into a more general multi-body control framework is part of our current efforts. This multi-body framework will make use of torque-control techniques, allowing for a robust and compliant interaction with the robot’s environment. The control framework will make use of multi-level hierarchies. Thereby, the robot will be able to do “multi-tasking” while in challenging situations all resources are available for the most critical tasks, such as balance.
Englsberger, Ott, and Albu-Schäffer: Three-dimensional bipedal walking control using Divergent Component of Motion. In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2013, Tokyo, Japan.
The development of TORO was supported by the Initiative and Networking Fund of the Helmholtz Association (HGF) via the Young Investigators research Group NG808.