Proposal Number: 0705025

Panel Summary: Panel Summary

I. A brief statement of what the proposal is about.

The focus of this project is to develop a small bio-inspired robot with an integrated vision system that can traverse difficult terrain. The vision system combines both stereo vision and monocular vision for depth estimation. Tactile sensing is also proposed to provide additional information on the terrain surface properties. The goal is to negotiate more difficult terrain than is currently possible by making advances in bio-inspired robots, perception of natural environments for legged mobility, and integrated planning and control over short ranges.

II. Intellectual merit - the innovative / driving research (strengths and weaknesses)

Strengths:

The problem of intelligent legged locomotion over difficult terrain is an important one, and a good team has been assembled to address the issues. The integration of a vision system with stereo and monocular vision for depth estimation is a good approach. Further, this is to be closely integrated with the motion planning and control. Using tactile sensing to provide additional information on the terrain surface properties should increase the robustness of the system.

Weaknesses:

The PIs propose stable locomotion over small obstacles, but they do not give any outline of the motion planning and control algorithms to accomplish this task both stably and robustly. They propose to choose foothold locations, but give few details on how to determine these from acceptable regions of support. More details are needed for the leg design to achieve greater adaptability than their previous legged systems -- underactuated?, degrees of freedom for general foot placement?, control of foot placement and contact force? A number of other issues have not been sufficiently addressed and these include the gait pattern, force distribution to avoid slip, etc., and dynamic stability. They propose to use tactile sensing to detect the terrain properties, but they don't provide details on the approach. The description of the new quadrupedal platform is quite weak. Design details are needed including its mechanism, sensor layout, and actuation scheme. More discussion on the proposed computer vision work is needed to elaborate on its novel aspects. How this work fits into the context of current work (DARPA, etc.) needs to be elaborated on to avoid overlap.

III. Broader impacts, including enhancing diversity and integration of research and education, if applicable (strengths and weaknesses)

Strengths:

The work has broad applicability to locomotion over unprepared surfaces including applications in search and rescue, planetary and lunar exploration, scientific exploration, and de-mining. The PIs have a history of involvement with undergraduate summer researchers and group presentations to high school students, and these will continue. The proposed work would fund one woman and two underrepresented students.

Weaknesses:

The PIs may want to consider incorporation of their results into coursework.

IV. Impact of prior NSF funded research, if applicable.

One of the PIs was funded for work on supervised dexterous telemanipulation with haptic feedback. The work resulted in new haptic devices, control algorithms, and improved understanding of shared control. Other work has investigated the skin stretch modality for haptic display. Hardware for the Stanford AI Robot (STAIR) has been funded, and the project has already produced several publications.

V. Panel recommendation and justification, including key strengths and critical weaknesses; also indicate if the project is considered transformative (high risk/high payoff) and why.

The panel places this proposal in the fund if possible category. The integration of vision on a small bio-inspired robot is clearly a good direction. It will extend the investigators' excellent work in legged robot mechanics and machine learning to address some very difficult problems of robust locomotion over uneven terrain.

The proposed work will be strengthened if several other issues are addressed. The need for a new platform from those developed previously is only weakly motivated. What new leg design is needed? More details on the interplay between the vision system, with its supervised learning strategies, and the robot mechanics could make the proposal more novel. Many challenges for robust and stable locomotion of a quadruped over unprepared surfaces have not been sufficiently addressed in the proposal.

The summary was read by the panel, and the panel concurred that the summary accurately reflects the panel discussion.

Panel Recommendation: Fund If Possible

Review #1

Rating: Very Good

REVIEW:

What is the intellectual merit of the proposed activity?

The proposed research of creating a legged bio-inspired robot capable of maneuvering the real world autonomously can be very challenging and a milestone for developing robust intelligent robot. The main goal of the proposal is to redesign the PIs' existing bio-inspired robotic platforms to include capability of maneuvering unstructured terrain without human intervention. To achieve this challenging research goal, the PIs propose an approach of sensing 3D terrain geometry and surface characteristics, and integrating them for robust control strategy. However, the design details of the new platform are not elaborated in the proposal, including its mechanism, sensor layout, and actuation scheme. Even if the PIs have a strong track record of building similar robots, providing more detailed robot mechanism that can implement the proposed locomotion strategy can make the proposal more compelling. The proposal includes a well-conceived research plan that would develop a bio-inspired robot platform over a series of experimental tests and both PIs have strong track record in the research area and are well qualified to conduct the proposed research.

What are the broader impacts of the proposed activity?

The proposed research can have a broader scientific impact in robotics research community and has many important future applications including surveillance and search and rescue, etc. Considering the PI's successful exposure to media for similar robot demonstration, if successful, the proposed intelligent robot can attract more attention from broader communities. The proposal has an adequate educational component from K-12, undergraduate, and graduate program, however, it lacks any planned involvement of minority students and no evidenced intent to incorporate the results in coursework.

Summary Statement

The PIs propose to develop and test biologically-inspired small-scale legged robots for natural unstructured terrain. The work builds on existing robot platform as well as on the newly designed platform with added capabilities of sensing and learning for autonomous maneuvering in natural terrains.

Review #2

Rating: Good

REVIEW:

What is the intellectual merit of the proposed activity?

This team of Stanford researchers proposes work that brings 3D terrain interpretation, learning algorithms and robotic perception together to create legged robots that traverse rough terrain. The envisioned end result is a robot that autonomously hikes a mountain trail over various terrain types.

The proposal cites that the critical gap in the knowledge domain is the lack of planning strategies that are closely tied to the robot's mechanics. Specific work includes designing and fabricating a new quadrapedal platform that integrates pan-tilt vision and geometric models of the terrain environment.

Intellectual Merit:

The team has a strong continuum of research. Their work and preliminary data suggests that filling the cited gap will vertically advance that field. The proposed work appears to overlap to a large degree with the DARPA programs (e.g. Biodynotics and LAGR) û which the team is currently performing or recently concluded. Indeed work that ties planning strategies with a robot's mechanics is intriguing. However, the proposal details the computer vision approach (not clear what is new here) for depth estimation and supervised learning strategies. More detail on the interplay such strategies have with robot mechanics could make the proposal more novel.

What are the broader impacts of the proposed activity?

If the proposed work on small legged robots fulfills its goals, it can broadly impact missions like search and rescue, disaster mitigation and exploration. In regards to outreach, the team has an established track record in training (research) students and promoting robotics (e.g. FIRST, media, high-school presentations, and lab tours). The team also has experience recruiting and training under-represented groups.

The above describes what has been done in the past. Unclear is how these past outreach activities relate to new ones. How does the proposed work effect FIRST robotics? What are the recruitment (and training and post-graduate) plans for integrating under-represented groups in the current proposal? Detailing these activities would strengthen the proposal. As it currently described, the outreach efforts, while laudable, are common to most university outreach and open house activities.

Summary Statement

Review #3

Rating: Very Good

REVIEW:

What is the intellectual merit of the proposed activity?

The proposed work is to build a next-generation quadrupedal platform with sufficient sensing and learning capabilities to allow it to operate autonomously in rugged terrain for long periods of time. The work is a collaboration of Cutkosky, a leader in biomimetic walking locomotion, and Ng, who has done excellent work in real-time learning control for dynamic robotic systems. The collaboration seems appropriate.

The robot will have a pan-tilt vision system for stereo and monocular vision along with other proprioceptive and tactile sensors. It will gather information about its environment (e.g., depth and terrain map, surface friction, etc.) and perform multi-step path planning. It will extend Ng's past work on apprenticeship learning and self-supervised learning.

The description of the new quadrupedal platform is quite weak. Is a new platform necessary? What weaknesses of existing platforms will be addressed, and how? What will be the size of the new robot? Why compare to Stickybot (e.g., "the new platform will have more ground clearance than Stickybot..."), which was designed for very different environments? I assume there will be no tether for power or other, as vision data will be sent wirelessly to an off-board computer.

What are the broader impacts of the proposed activity?

Legged robots that are able to operate autonomously (or semi-autonomously) will have a great variety of applications in search-and-rescue, military scenarios, etc. Past work by the PIs has resulted in outreach by significant popular media coverage. The PIs also have a strong record of high school outreach. The proposed work would fund a woman and two students from underrepresented groups.

Summary Statement

Review #4

Rating: Very Good

REVIEW:

What is the intellectual merit of the proposed activity?

The focus of this project is to develop a small bio-inspired robot with an integrated vision system that can traverse difficult terrain. The vision system combines both stereo vision and monocular vision for depth estimation. Tactile sensing is also proposed to provide additional information on the terrain surface properties. The goal is to negotiate more difficult terrain than is currently possible by making advances in bio-inspired robots, perception of natural environments for legged mobility, and integrated planning and control over short ranges.

The perception system will be integrated with the motion planning and control system to estimate terrain properties to establish good footholds. The perceptual models of the world will be updated through tactile exploration as the robot moves over the terrain. The challenge for the system is to autonomously traverse a mountainous hiking trail.

What are the broader impacts of the proposed activity?

The PIs outline several points of broader impact to the technical community, educational community, and community at large. The major scientific contribution will be to advance the use of legged vehicles in off-the-road applications. With success in this area, legged vehicles may find application in search and rescue, planetary and lunar exploration, geological and volcanic exploration, and de-mining.

The PIs and their research labs have a good history of outreach to the public through science-education programming. They actively present their work to high school students and recruit undergraduate researchers. Of their current Ph.D. students, one is female and two are from underrepresented groups.

Summary Statement

The problem of intelligent legged locomotion over difficult terrain is a very important one. Clearly, advances are needed in the mechanics and control of legged vehicles, and in the integrated perception of the terrain. The PIs outline a very ambitious plan to make advances in each of these areas. The approach to combine stereo and monocular vision to determine footholds appears to be a good one. However, it is not clear that they have a viable plan to investigate surface roughness, etc. with tactile sensing. It is an important problem, but may need a much more comprehensive investigation in its own right.

Previous work by one of the PIs has focused on underactuated hexapods. The results have been impressive in terms of their robustness. However, in order to adapt to more difficult 3D terrain by careful foothold selection, the motion planning and control system must be much more sophisticated. Also, stability for a quadruped, with the controlled 3D motion capabilities needed, is not easily managed. The PIs have not adequately addressed this problem in the proposal. It is key to adaptive locomotion on unprepared terrain. It is not clear that they will develop anything more than a robot that traverses small obstacles by "flailing at them". The leg coordination and control problem is critical to this success. Also, the 3D motion and force capabilities of individual legs may require a more sophisticated leg design, and details of their design to address this problem are needed in the proposal.

-- AlanAsbeck - 09 Aug 2007