Research

 

Motivation

    The methodology of locomotion has undergone dramatic change with the advancement of human civilization. During the last two centuries, various remarkable modern transportation methods, including wheeled and track vehicles, airplanes, were invented based on the discovered physical principles and designed to adapt to a specific environment. However, development of vehicles that are capable of passage over natural rough terrains remains primitive. Are humans sufficiently clever to invent a new machine to serve this purpose using presently known design theories and methodologies? Perhaps, but I am more inclined to believe that a systematic and incremental approach building on learning the lessons from nature offers a more likely path.

    Animals provide an ideal inspiration to the legged robotic world. After millions of years of evolution, animals have developed high complex degree-of-freedom (DOF) systems yet perform great mobility that is unparallel to the existing robotic machines. Though animals’ appearances and structures may vary significantly, for similar behaviors, for example, running, they seem to share a few similar fundamental motion patterns which can be represented by simple low DOF spring-damper-mass models (“templates”). Recent-developed bio-inspired legged machines such as RHex and iSprawl whose running motion can be represented in the similar model as in animals indeed have begun to exhibit a much greater mobility. Of course nature has its own mechanism for evolution and formats animals into high DOF systems after generations---reflecting most likely the necessity of performing a wide diversity of behaviors to ensure survival? Probably true. No matter what the answer is, this concept opens a new challenge yet fundamental question: How to systematically design and control a robot (or a general physical dynamic system) based on a bio-inspired model or even multi-models? I believe (1) systematically learning from the biological design and behavior offers solid concept and intuition of dynamic locomotion. (2) Analyzing a particular behavior by assigning it a distinguished, simple model opens the gateway to truly understand the essence of design and control of each particular constituent of the larger behavioral suite. (3) With right composition of multi templates where each presents a unique behavior, this approach to research promises the capability in the long run to design and control a much more complex high-mobility system capable of performing multi agile behaviors and responding to diverse nature environments.

    I have no doubt that someday the legged machines will become one of human’s major transportation methods as “car” was invented as wheeled human-carrier 100 years ago. However, it is a challenge task even for a small-size robust and agile legged machine to achieve the modest capabilities that would entitle use to call it a “robot”.

 

Research Work

    My research activities and interests lie in the understanding/improvement of performance/behavior of bio-inspired legged machines (short-term) and general physical dynamic systems (long-term) from two aspects: design and control. Design with the appropriate morphology and materials (“form”) forms the substrate on which system performance can be elicited by appropriately matched control (“function”). In my graduate and post-doctoral research I have worked with diverse teams on various portions, detailed in the summaries listed below, of this long-term task as depicted in this figure:

 

Design of a Bio-inspired Dynamical Vertical Climbing Robot

 

Movies:

  Front view

  Side view

This paper reviews a template for dynamical climbing originating in biology, explores its hypothetical utility , and offers a preliminary look at empirical data bearing on the feasibility of adapting it to build a robot that “runs” vertically upward. The recently proposed pendulous climbing model abstracts remarkable similarities in dynamic wall scaling behavior exhibited by radically different animal species. The present paper’s first contribution summarizes a continuing numerical study of this model to hypothesize that these animals’ apparently “wasteful” commitments to lateral oscillations may be justified by a significant gain in the dynamical stability and, hence, the robustness of their resulting climbing capability. We explore numerically a scaled version of this template devised to inform the design of a physically realizable robotic mechanism with the same climbing behavior. The paper’s second contribution documents the design and offers very preliminary empirical data arising from a physical instantiation of this model. Notwithstanding the significant differences between the proposed bio-inspired template and its physical robot model, these intial data suggest the mechanical climber may be capable of roughly reproducing both the motions and ground reaction forces characteristic of dynamical climbing animals. Even without proper tuning the robot’s steady state trajectories manifest a substantial exchange of kinetic and potential energy, resulting in vertical speeds of 0.14 m/s (0.35 bl/s) and claiming its place as the first bio-inspired dynamical legged climbing platform.

(in collaboration with Dr. Jonathan Clark)

 

Spontaneous formation of 1D ripples in transit to highly-ordered 2D herringbone structures through sequential and unequal 2D mechanical force

Spontaneous formation of periodic structures with controlled morphologies on surfaces has been of great interest for many potential applications. We report the formation of various submicron wrinkle patterns using mechanical force coupled with oxygen plasma treatment on PDMS. It allows us to control the amount and timing of strain applied to the substrate on both planar directions (either simultaneously or sequentially), which appears to be critical to maneuver the pattern formation of 1D ripple, 2D herringbone, and patterns in between in real time. We observe clear transitions from ripple, to ripple with bifurcation, to ripple/herringbone mixed features, and to completely 2D herringbone structure. More specifically, we demonstrate the well-controlled formation of a highly-ordered zigzag-based herringbone structure, and elucidate the mechanisms of pattern formation and transition at a large strain level (up to 60%).

 

Distributed Mechanical Feedback in Arthropods and Robots Simplifies Control of Rapid Running on Challenging Terrain

 

some movies

Terrestrial arthropods negotiate demanding terrain more effectively than any search-and-rescue robot. Slow, precise stepping using distributed neural feedback is one strategy for dealing with challenging terrain. Alternatively, arthropods could simplify control on demanding surfaces by rapid running that uses kinetic energy to bridge gaps between footholds. We demonstrate that this is achieved using distributed mechanical feedback, resulting from passive contacts along legs positioned by pre-programmed trajectories favorable to their attachment mechanisms. We used wire-mesh experimental surfaces to determine how a decrease in foothold probability affects speed and stability. Spiders and insects attained high running speeds on simulated terrain with 90% of the surface contact area removed. Cockroaches maintained high speeds even with their tarsi ablated, by generating horizontally oriented leg trajectories. Spiders with more vertically directed leg placement used leg spines, which resulted in more effective distributed contact by interlocking with asperities during leg extension, but collapsing during flexion, preventing entanglement. Ghost crabs, which naturally lack leg spines, showed increased mobility on wire mesh after the addition of artificial, collapsible spines. A bioinspired robot, RHex, was redesigned to maximize effective distributed leg contact, by changing leg orientation and adding directional spines. These changes improved RHex’s agility on challenging surfaces without adding sensors or changing the control system.

(in collaboration with Dr. J. C. Spagna, Prof. D. I. Goldman,  and Prof. R. J. Full at UC - Berkeley)

 

RHex-SLIP: A Model of the Robotic Hexapod RHex in the Sagittal Plane

The spring-loaded inverted pendulum (SLIP) is a simple, passively-elastic two-degree-of-freedom model for legged locomotion that describes the saggital-plane center of mass (COM) dynamics of many animal species and some legged robots. In previous work we have extended SLIP to model three-dimensional COM motions and to incorporate multiple stance legs. To better understand the agile hexapedal robot RHex, here we incorporate key details of leg design and motor specifications into SLIP, allowing us to match SLIP gaits with experimental data from RHex, and to investigate their stability properties. We find that motor and leg characteristics, and leg touchdown and liftoff protocols, can significantly influence stability, and that non-periodic "chaotic" gaits can occur.

(in collaboration with Dr. J. Seipel and Prof. P. Holmes at Princeton University)

 

Sensor Data Fusion for Body State Estimation in a Hexapod Robot with Dynamical Gaits

We report on a hybrid 12 dimensional full body state estimator for a hexapod robot executing a jogging gait in steady state on level terrain with regularly alternating ground contact and aerial phases of motion. We use a repeating sequence of continuous time dynamical models that are switched in and out of an Extended Kalman Filter to fuse measurements from a novel leg pose sensor and inertial sensors. We implement this estimation procedure offline, using data extracted from numerous repeated runs of the hexapod robot RHex (bearing the appropriate sensor suite) and evaluate its performance with reference to a visual ground truth measurement system, comparing as well, the relative performance of different fusion approaches implemented via different model sequences.

 

Advanced Inertia Measurement Unit (IMU) --- with 12-Axis Accelerometer Suite

This inertial measurement unit supplements the traditionally paired three-axis rate gyro and three-axis accelerometer with a set of three additional three-axis accelerometer suites, thereby providing additional angular acceleration measurement, avoiding the need for localization of the accelerometer at the center of mass on the robot's body, and simplifying installation and calibration.

 

Legged Odometry from Body Pose in a Hexapod Robot

We report on a continuous time odometry scheme for a walking hexapod robot built upon a previously developed leg-strain based body pose estimator. We implement this estimation procedure and odometry scheme on the robot RHex and evaluate its performance at widely varying speeds and over different ground conditions by means of a 6 degree of freedom vision based ground truth measurement system (GTMS). We also compare the performance to that of sensorless odometry schemes --- both legged as well as on a wheeled version of the robot --- using GTMS measurements of elapsed distance.

 

A Leg Configuration Sensory System for Dynamical Body Pose Estimation

We report on a novel leg strain sensory system for the autonomous robot RHex implemented upon a cheap, high performance local wireless network. We introduce a model for RHex's 4-bar legs relating leg strain to leg kinematic configuration in the body coordinate frame. We compare against ground truth measurement the performance of the model operating on real-time leg strain data generated under completely realistic operating conditions. We introduce an algorithm for computing six degree of freedom body posture measurements in world frame coordinates from the outputs of the six leg configuration models, together with a priori information about the ground. We discuss the manner in which such stance phase configuration estimates will be fused with other sensory data to develop the continuous time full body state estimates for RHex.

 

A Context-Based State Estimation Technique for Hybrid Systems

This paper proposes an approach to robust state estimation for mobile robots with intermittent dynamics. The approach consists of identifying the robot’s mode of operation by classifying the output of onboard sensors into mode-specific contexts. The underlying technique seeks to efficiently use available sensor information to enable accurate, high-bandwidth mode identification. Context classification is combined with multiple-model filtering in order to significantly improve the accuracy of state estimates for hybrid systems. This approach is validated in simulation and shown experimentally to produce accurate estimates on a jogging robot using low-cost sensors.

(in collaboration with Dr. S. Skaff, Prof. A. Rizzi, and Prof. H. Choset at Carnegie Mellon University)