2019-06-19: Ai's Latest Demo for Multiple Agencies

Several representatives offered some interesting questions and points
relevant to our research.

Some of the most exciting moments here at the Autonomy Incubator include when we get to offer demonstrations to curious individuals and groups. This week, representatives from several different government agencies including NRL and AFRL, visited to take a tour of NASA Langley, and stopped by our branch.

We had the pleasure of showing them our flight room, which is still in the works of being renovated (and we can't wait to start using it again soon)!

Danette Allen showing off our flight room and wire maze.

Danette Allen, Walter Waltz, Sherif Shazly, and Ben Hargis each got to present a little bit about their research, speaking on behalf of their recent accomplishments and goals.

Since our main focus at the Ai includes multi-agent collaborative autonomous assembly, the team discussed what we have accomplished to date with the In-Space Assembly (ISA) project through a sub-scale demonstration.  For this demonstration, we worked with multiple quarter-scale trusses and a single robotic arm.

We do have to do a lot of research via simulation, and I had the pleasure of constructing a few videos so our team members could demonstrate what that looks like.

Danette walking through one of the videos.

Danette passed the baton to Walter, so he could share what he has been working on for the last year.  He described his research with single-agent autonomous assembly, where he focuses on motion planning so that the robot can successfully undertake each step of the assembly process. The four major aspects that his work has an emphasis on are object detection, motion planning, collision avoidance, and trajectory execution.

Walter Waltz has been at the Ai for about a year.

The robotic arm will pick up a truss from a random location and place it in a predefined location.  As it is going through the mission, there is a dynamic sequence of different stages that it will go through, so that he and the other researchers can rigorously test new algorithms, see different planners, and so on.

For these autonomous behaviors, it is extremely important to consider the motion planning and execution. One of the most unique elements about some of the trajectories is that they will observe constraints, which will slightly complicate the planning and execution. 

Of course, all of this development begins in a simulation environment, as previously established. This is so Walter and his research team can get a better idea of the algorithms and how they can form requirements, leading towards validation and the rigorous classification that is necessary for flying to space.

Sherif is one of the robotics software developers on the ISA team.

Following Walter, Sherif took the spotlight to discuss his focus on using 3D point clouds to generate occupancy grids of the collision scene of the robot's workspace. 

The algorithm encodes information about whether the subspace is occupied, unoccupied or unexplored, and by using that information, it creates plans to avoid not only hitting the truss assembly but also any object.  It is very efficient in doing so, which allows us to share the planning scene efficiently as well. This is very important for the ISA researchers because we, as a team, have limited resources, and we do not want to waste what we have and can get.  We are hoping to expand this idea through the inertial transfer concept, which was further detailed by Ben.

This is Ben's second summer with the Ai as a Pathways intern.

Ben defined inertial transfer as "the concept of moving untethered objects through space using the object's own inertia."

Through the use of a concept video from the RAMSES project, he was able to give a detailed description of how it works.  First, the front manipulator arm transfers an instrument panel from storage to the manipulator that is in charge of installing the panel. The efficiency here is gained by free-flying mass, and there are additional structures that allow these manipulators to travel the distance shown in the graphic.  As soon as this object leaves the grasp of the manipulator, it can thus be tracked depending on what metrology strategy is employed.

As a relatively new project, our exploration unveils some questions that need to be answered, such as what kind of infrastructure is best for communication, what information needs to be passed and when, and what are our metrology strategies are- what are we going to measure?

All of these steps are very important in mitigating risks so that we can learn from each consideration. There is always going to be some randomness to the process, which is why it is important to have contingency plans for failed grasps. This can range from anything like awkwardly grasping an object and ending up at an impasse, or a near miss.

A major goal, of course, is to reduce and mitigate as much of this risk as possible.

It was the Ai's pleasure to present another demo and answer any questions that our visitor's asked.  They also made some very useful points for us to consider in the future.

Thank you ISA community, we hope to see you again!