C.1 Source Acquisition


One of the more prominent areas where animals outperform robots is their ability to react and adapt to unexpected or atypical circumstances. My initial thought for this assignment was to investigate how animals diminish the impact from a fall or a jump, particularly larger animals. I think this was in part prompted by noticing the harness on the class demo of the RiSE robot; I was thinking about how valuable it would be to have a robot that didn't break any time it took a tumble.

I didn't have a great idea of how to phrase this, so I started a bit naively, first searching scopus for:

biomechanics breaking a fall, then
biomechanics fall arrest (found the phrase "arresting a fall")
biomechanics fall arrest -hand (since most articles were about upper extremities)
biomechanics jump landing (what I wanted more was something along the lines of quadrupedal joint flexion, which would likely come into play in a jump)
biomechanics quadrupedal jump landing

Most of what I was coming up with was essentially how humans deal with tripping, which was interesting but I wanted a more holistic treatment of falling and jumping. So I sat back to think a bit more about falling and landing behaviors observed in animals, and identified four main sources of bioinspiration:

  1. Humans absorbing minor impacts via
    1. Leg flexion for jumping
    2. Arm flexion for falls from standing
  2. Humans lessening forceful falls or jumps by doing a forward roll (e.g., parkour)
  3. Quadrupeds absorbing impact via flexion of 4 legs
  4. Roaches and other insects surviving falls by merely being light and durable (I believe)

Of course, 4) would likely not apply to most larger robots. So for a decent-sized robot to handle falling, it would have to incorporate aspects of 1), 2), and 3). Thinking further, the landing itself is not the full picture— there are at least two other necessary capabilities:

1) The robot must identify when it is falling and what kind of fall is happening (e.g., backward from standing, jumping from 10 ft, etc.) and initiate the appropriate response.

2) The robot must then position itself in such a way as to be prepared for the identified landing (leg flexion, rolling, etc.); programmed landing behaviors are useless if the robot cannot position itself to execute them effectively. For example, if a humanoid robot falls over sideways and doesn't twist its upper body there is no opportunity to lessen the impact using arm flexion.


I therefore saw essentially three broad components to handling a fall: recognition and decision making, preparation for landing, and execution of landing behavior. Each seemed to be a rich topic, and I thought it unlikely that any one paper would adequately handle all three. I decided that it would make sense to focus on one of them, and began to survey the literature before choosing one to look at in more depth.

As for landing behavior, I thought it would be quite nice if a robot could execute a forward roll to turn vertical momentum into horizontal momentum on landing (thus minimizing force of impact). However, I was not able to find anything relevant after a cursory literature search, and frankly I doubt the field is advanced enough for anyone to be seriously investigating this type of behavior (there was a robot that could do a somersault, but it was not dynamic at all). Aside from rolling, I wasn't terribly interested in focusing on joint flexion, so I decided to move on to fall processing.

This area of inquiry was a bit more fruitful, and I found several papers addressing the topic. Most were concerned with humanoid robots in particular (several aimed at soccer robots); one identified their tendency to "fall over and then damage severely" to be "one of the crucial barriers for practical application of humanoid robots."

While bipedal fall management would certainly be worthy of this assignment, I really was interested in the ability to safely handle falls or jumps from non-negligible heights such as those encountered by a climbing robot. Being able to height, velocity, and distance to impact are essential components of this, as well as having the capacity to absorb shock on impact. However, I felt the more interesting problem was landing preparation— the robot must be oriented in such a way as to take advantage of its shock absorbing components (if you drop a pogo stick on its side, it just crashes).

So I now had my question: How do animals prepare for landing in response to a jump or fall, and how can that capability be implemented in robots?

Literature search

I was inspired by the ability of cats to always land feet first, even when dropped from upside down. I hadn't thought too much about how they did it, but had always been a bit puzzled by the phenomenon since it seemed to violate conservation of angular momentum. So I decided to search the databases for papers about falling cats and robotics.


A Scopus search for "cat robot*" sorted by citation only returned three relevant papers in the first 50 results: [1], [2], and [3], the last two of which are in Japanese and all of which are from the 1990s. However, the first had 75 Scopus citations, making it a good candidate as a "seed" for future searches. Indeed, it introduced me to the keyword nonholonomic, which describes the type of motion experienced by a falling cat.


In addition to turning up the results above, ISI also produced [4], which is from the highest ranked robotics journal according to the ISI Web of Knowledge, but only has 5 cites in the last 9 years.

Google Scholar

"cat robot" produced a fairly unfocused set of results, but "cat robot nonholonomic" was more on target, and limiting the papers to the past five years I identified three conference articles whose topic looked promising: [5], [6], and [7]. Note that the first two are the same authors and general topic at different conferences, and there was a third by them as well.


Research into the falling motion of cats has been conducted [8] since the late 19th century. It seems that it progressed from high-speed photography in 1894 [10] to a 1936 mathematical model [9], to a 1969 refinement [11], and to the 1994 paper found previously via Scopus [1].


C.1.0) Source

I chose to use reference [7] due to the fact that it is highly relevant to my topic of research, it comes from a reputable venue, the authors are from reputable institutions, and even as a recent paper it already has a citation in a high-impact robotics journal.

C.1.1) Venue

IEEE conferences are generally a credible source due to the credibility of IEEE itself, and the theme of the conference was "Exploring New Horizons in Intelligent Robots and Systems", which is certainly an appropriate theme for this paper.

C.1.2) Authors' Qualifications

Mark Yim is a professor at the University of Pennsylvania, a credible academic institution, and Thomas Mather is one of his students. Yim has a respectable h-index of 13, particular considering that Scopus tends to undercalculate the h-index.

C.1.3) Source Identification Methods

I was unable to access journal articles at first, but Doug was able to help me with access to IEEE Xplore.

C.1.4) High Quality Bioinspired Robotics Contribution

This explores an new implementation of nonholonomic aerial movement, which lends itself to further research and application.

C.1.5) General Robotics Literature

The other papers cited from the Google Scholar search ([5] and [6]) also demonstrate an attempt at solving the posed question. A recent paper [12] citing [7] takes a different approach to the problem, inspired by the lizard tail, and is published in Bioinspiration and Biomimetics, ranked 3 in robotics journals according to the ISI Web of Knowledge. Both acknowledge the problem and present partial solutions to it.

C.1.6) Biology Literature

The references previously cited under "Biology" are established and well-cited biology literature: [9], [10], [11], and [1]. Each implicitly acknowledge the utility of cats' capability to rotate mid-air.