A robot bodies balancers : solve walking probelems

Daan Hobbelen engineer at the University of Delft, developed the robot Flame consists of seven devices, an organ of balance. Several algorithms allow it to find and keep balance. Flame can perform 20 to 25 no, while his predecessor, Meta, could make almost a hundred until it reaches a wall. Meta could also overcome obstacles and make no 3 cm high, while humans typically rise 5 to 6 cm. Hobbelen believes that Flame will be able to walk that way in a few months and may move more laterally.

“Walking is very complex. It is a natural exercise for human beings, but we do not know really how it works,” says Hobbelen. This research topic was chosen in the first place to help people with walking problems. The results obtained with Flame, Hobbelen can develop better diagnostics, as well as meetings and instruments more effective rehabilitation. “By building this robot, we have new data that may help develop new prostheses, because it is impossible to know and measure how the human being is in balance.” Hobbelen also hopes that this type of robot could play a role in home care, especially among the elderly. The entertainment industry may also be an important market, according Hobbelen.

DBL’s research background

The Delft Bio-robotics Laboratory was started in 1995 with Van der Linde’s walking robot research. The walking robot research aims at the development of human-like walking machines. This project is a clear example of biologically inspired robot design; instead of jerky robotic motions we want to achieve a natural walking pattern. We expect that our research will provide insight in the fundamental principles of human walking, without distraction by the enormous complexity of the multi-purpose machinery of the human body. This insight, in turn, we hope can help the development of new rehabilitation devices. Also, the entertainment business could well benefit from a better understanding of how to develop walking machines.

Theapproach to walking machine design originates from the prosthetics design practice within the MMSC-group. Van der Linde was dared by his master’s thesis advisor, emeritus prof. Cool (pronounce Cole), to build a walking machine powered by only one small motor. Minimal weight and energy consumption are important design criteria for prosthetics, and could shed new light on machine walking. Van der Linde succeeded with his waddling Stappo.

van der Linde

Soon we found out not to be the first with this approach by far. Official recordings of walking toys date back as far as 1888 (e.g. Fallis patent). In 1989, McGeer started a systematic research into “Passive dynamic walking machines”, which walk down a shallow slope and need no motors and no controls. His research was continued by prof. Ruina in his “Human Power and Robotics Laboratory”. Wisse, then an undergraduate student, visited Ruina’s lab in 1998 to build a 3D walker. Steve Collins and prof. Ruina completed the machine.

Prof. McGeer

Prof. Ruina

We regard passive dynamic walking as the starting point for all our walking machine designs, even though we are powering some of our machines such that they are not purely passive. We apply two concurrent research strategies. On one hand, we are looking into irreducibly simple, idealized walking models, to discover general truths about bipedal (two-legged) walking. Next to the simple, idealized models, we are developing real-world walking prototypes. With these prototypes we want to test the fundamental principles discovered with the idealized computer models. The prototypes serve as proof of applicability, and also as a source of new ideas.

In chronological order, we have developed:
Stappo, Van der Linde’s first walking (waddling) machine,
The Cornell walker, which was completed in Ruina’s lab,
Bob, a 3D walker with knees and ankles, which was just a little too underactuated,
Baps, the much more versatile successor of Stappo, and the first to use pneumatic McKibben muscles,
The Museon walker, a conceptually simple but eye-pleasing demonstration machine,
Mike, a 2D machine with knees, actuated by McKibben muscles,
Max, a 2D walker with knees, hips and an upper body.

More about Flame-


Flame is our first fully 3D walking robot with electric actuation. Similar as in Meta, Maxon DC motors are used to actuate several joints: two sagittal ankle joints, two sagittal knee joints, two sagittal hip joints and one lateral hip joint (for sideways foot placement). All the actuated joints in the sagittal plane are implemented with series elastic elements that allow the application of torque control. Next to the actuation, all joint are equipped with incremental encoders that measure all joint angles and indirectly the elongation of the elastic elements.


The upper body holds most of the essential electronics present in Flame. A PC104 computer (with all the required I/O onboard) has a real-time Linux kernel (RTAI) that runs a control loop at a sample rate of 1 kHz. The control loop implements torque and/or position control on all actuated joints depending on the task at hand. Next to this central computer, important electronic components that are found on the upper body are the inertial sensor (from XSens) and the current amplifiers.

Overall, Flame weighs approximately 15 kg and is about 1.3 m tall. At the time of writing it walks at a speed of 0.45 m/s and is able to handle stepdown disturbances up to 8 mm. Continuing research is aiming at a wide speed range with both slower and faster speeds, starting and stopping, as well as larger disturbance handling. The main research topic that has been covered with Flame is the synthesis of 3D stable gait with the application of a (simple) sideways foot placement strategy using information from the inertial sensor.