Fast and robust two-legged walking robots


Update content

Title Fast and robust two-legged walking robots
Period 11 / 2006 - 11 / 2009
Status Completed
Research number OND1318882
Data Supplier Website STW


Research content The current level of technology enables the realization of a long standing dream: building a robot that walks like a human. Companies in Japan (Honda, Sony) are well-known for their construction of human-like robots. These robots are chock-full of high-grade technological components - yet the control is based on a conventional approach (i.e. strong motors follow preprogrammed trajectories using high controller gains). In a sense, the ready availability of highgrade components brings a huge drawback; it hampers theoretical innovation and one could lose sight of the real goals of biped research. Eventually, the research has to lead to (1) practical, affordable assistant robots, and (2) applicable knowledge on human walking for rehabilitation. These goals will not be met with the (non-human) conventional control approach requiring expensive components. Here we propose an unconventional approach labeled "limit cycle control" which will lead to simple, efficient, robust, and fast walking devices. The main problem with conventional stability control is that the step-to-step transitions are incorrectly regarded as disturbances. The walking motion is regarded as a succession of singlefooted balancing-exercises interrupted by destabilizing step-to-step transitions. The faster the walking motion, the more frequent such destabilizing transitions occur. This explains why conventional robots excel in standing still and in slow walking, while having difficulty walking at normal and fast velocities. Our research shows that the conventional approach is incorrect or at least incomplete. The transitions can actually help stability. Moreover, we have built passive dynamic robots that walk without any conventional control, obtaining their stability completely from the step-to-step transitions. Our most recent prototype was published in Science (February 2005), commenting on the enormous potential of this approach; the robots are not only elegantly simple, but also show efficient and natural walking motions. The core new insight of this proposal is captured in the term "limit cycle control". The walking motion is better regarded as a periodic motion (limit cycle) which gets its stability thanks to the step-to-step transitions. An immediate pay-off ("low-hanging fruit") of this insight is the following: the higher the step frequency, the simpler it becomes to find stable walking motions. In other words, our robots are more robust when walking faster. In this project we will explore how this insight can be exploited by the mechanical design and by the control of the bipeds. We will research the contribution to robustness of the following ideas: ankle push-off, attitude control of the upper body, lateral foot placement and automatic optimization of control patterns. This list of ideas will increase over the duration of the project, as each new prototype brings new experience and new ideas. Successful locomotion research requires the interaction between simulation studies and prototype experiments. We intend to entertain a continuous interaction between the two by studying each specific aspect or feature both in a simulation and in a new or modified prototype. We expect that this will lead to a breakthrough development with a good chance that all future human-like robots will be based on limit cycle control. This will be an essential prerequisite for the development of affordable assistant robots. Moreover, we expect to finally understand how humans maintain their stability with such great ease, and how we can help people confronted with impairments. Utilization We belong to a small number of research groups that follows an unconventional philosophy, utilizing step-to-step transitions for stability rather than using conventional stability control. Within this small community, our innovation is to consider the beneficial influence of increasing the walking velocity. The goal is to develop two-legged robots that can walk stably on rough terrain a factor ten better than current prototypes, as measured by the roughness of the terrain that they can handle. With our new approach, the walkers will be twice as fast and many factors less complex and cheaper than current prototypes. Walking robot toys sell well, as demonstrated with the commercial success of RoboSapien (1.5 million sold worldwide). Our project will result in toys that will walk fast enough to be entertaining as a radio-controlled two-legged vehicle (a 20 cm tall toy is expected to walk at 3 km/h) while it will likely require only a single motor and two foot switches for stability control. We will commercialize the concept into a product by collaborating with Lithp Systems, a Dutch robot development company. It is expected that this can happen within the four-year project period. Simultaneously, robust, fast, and simple walking robots are interesting for new entertainment attractions, for which we collaborate with the Dutch theme park De Efteling. This development can be extended towards, eventually, fully autonomous assistant robots. This is the interest of participants TNO and Philips and component suppliers Maxon Motor and Bergmann. A huge market on household robotics is emerging (ranging from automated vacuum cleaners to human like assistant robots). This is mostly a new market with correspondingly new companies. If the Dutch society invests now in these emerging markets, there is strong chance of developing a significant industry in this field, similar to the strong Dutch position in the market for household appliances with Philips. The stability control approach is strongly related to human walking. For example, we may find an explanation for the counterintuitive findings in biomechanics studies that faster human gait seems to be more regular. The concept also suggests that it is possibly easier to learn to walk fast than to walk slow, explaining why many kids learn to walk a few steps (e.g. between the table and chair) before they learn how to stand still without help. The company Ambroise participates in order to implemen t the project results in novel prostheses and orthotic devices.

Related organisations

Related people


D14240 Manufacturing technology, mechanical technology, robotics

Go to page top
Go back to contents
Go back to site navigation