Design, modelling and control of slider: an ultra-lightweight, knee-less, low-cost bipedal walking robot

Abstract

Robots can improve the world we live in by replacing humans in dangerous and tedious tasks. More specifically, bipedal robots can navigate complex and unstructured environments designed for humans which may not be possible for their wheeled counterparts. This thesis presents a comprehensive study on the design, modelling, control and motion planning for a novel ultra-lightweight, knee-less, low-cost bipedal walking robot called SLIDER.

Although many bipedal robots exist, it is still a big challenge to design and build a bipedal robot that is lightweight, able to perform agile motions while keeping the cost low. The first part of the thesis aims to tackle this challenge by proposing a novel knee-less robot design. Unlike most bipedal robots which have knees, SLIDER has no knees. Instead it has prismatic sliding joints at the hip that replace the functionality of knees. This design results in lightweight legs which in turn bring advantages for agile control and improved energy efficiency. This thesis presents the details of the design and modelling of SLIDER, a comparison between knee-less and anthropomorphic bipedal robot design as well as real robot experiments. Furthermore, lessons learned through multiple iterations of the design and modelling of SLIDER are summarized.

Besides the design and modelling, the control and motion planning algorithms are another key part for achieving bipedal locomotion. The second half of the thesis presents contributions to the walking controller, footstep planning and dynamic motion generation for bipedal robots. Three algorithms are proposed. The first algorithm enables bipedal robots to blindly walk over uneven terrains by decoupling the aSLIP model and formulating the footstep planning as a linear MPC. The second proposed algorithm enables bipedal robots to adapt step location and timing by solving a nonlinear MPC. The nonlinear MPC can be solved with a frequency of 200Hz by an asynchronous framework. The third proposed algorithm generates highly dynamic jumping motions for bipedal and quadrupedal robots. A new model called LL-SRBM approximately models the centroidal inertia and improves the jumping performance in terms of jumping speed and distance.

The final contribution of the thesis is a novel proposed walking gait called RotoGait. Thanks to the unique design of SLIDER, the legs can continuously rotate while walking, instead of accelerating and decelerating each swing leg as in the normal gait. Trajectory optimization is used to generate the optimal energy-efficient swing leg trajectory for RotoGait. Compared with the normal walking gait, RotoGait proves to be more energy efficient at high walking speeds. RotoGait represents a new class of walking gaits that opens the possibility of improved locomotion performance.

Looking ahead, the novel insights shown in this thesis open up future research avenues for the design of bipedal robots and performance improvement of bipedal locomotion.