ARM-H
c. 2012
DARPA Autonomous Robotic Manipulation - Hardware Program
The ARM-H program was a program that “seeks to enable autonomous manipulation systems to surpass the performance level of remote manipulation systems that are controlled directly by a human operator. The program will attempt to reach this goal by developing software and hardware that enables robots to autonomously grasp and manipulate objects in unstructured environments, with humans providing only high-level direction.”
It was a multi-tracked program where “the hardware track focuses on design and development of low-cost dexterous multi-fingered hands taking advantage of recent manufacturing advancements.”
I was a part of one of the three projects attempting to win the down-select for the next phase of the program. Our project was a collaboration between Meka Robotics and SRI where Meka would provide general mechanical design, actuator design, electronics and control. SRI contributed a novel electro static clutch design used in the finger joints and expanded on some of our work to make tactile “skin” to provide pressure sensing across the fingers and palm.
The shape and finger layout were optimized for *ahem* handling… the tasks while minimizing the number of motors and overall weight. The fingers were also designed to be easily replaced in case they were destroyed. Almost all of the structure was sheet metal to reduce cost and all of the fingers were identical and interchangeable to simplify the BOM and assembly.
The actuators themselves were pretty simple and utilized cheap hobby BLDC motors and simple spur gears to wind up and pull a string that ran down the finger. The module included something we considered to be a pretty novel way to do force control of the figure grip. It used a simple loadcell commonly used in package scales to measure components of the tension in the string. This structure allowed the module to remain compact and provided fantastic cheap force control.
Each knuckle in a finger had position encoders and electrostatic clutches developed at SRI. With the encoders and clutches it was possible to lock the finger elements into specific configurations to execute specialized motions like pinching and rolling objects between the fingers.
Although we crushed the evaluation we were not selected for the next phase of the program. I think we actually performed all of the tasks faster and more effectively but likely we weren’t chosen because we were perceived to be more fragile.
The demo reel for the hands capabilities.
Demo of the tactile sensors. They were basically a grid of capacitive touch sensors on the bottom layer and a flex PCB to create a disturbance in the field like a human finger and sheet of urethan between to provide squish/spring.
Montage of our evaluation at the DARPA testing facility. The robot arm was teleoperated and I directly controlled the fingers with a little MIDI fader to set the forces and enable the clutches. It would be someone else’s job to figure out the autonomous motion planning if we were selected for the next phase.
The gear box was a pretty simple 2-stage spur gear setup. The bottom aluminum disk is the encoder for motor control. The aluminum disk in the center of the larger gear is the drum to wind up the string. The sting snakes through the two pullies above the drum out a small hole in the module. The center pully is mounted to a cantilevered loadcell (silver bar thing along the right). Because the pullies are offset, when the string tension increases it pulls the loadcell pully toward the center of the module (the left in this picture). Through some simple math and assumptions we can use that force to estimate the tension in the string.
The naked finger hardware! The silver lines dangling down are the wires for the high-voltage clutches located in each knuckle. You can also see the elastic urethan return spring hanging down below the finger (the tan rubber-band looking thing). The return spring pulls the finger straight when there’s no tension in the actuator string.
Here are some photos the evaluation day the the DARPA facility. This image gives a good sense of how the robot arm was teleoperated with copy of the robot arm (I’m holding onto it with my right hand). I controlled the fingers with a little MIDI fader to set the forces and enable the clutches with my left hand.
There were a lot of different tasks ranging from picking up a battery and dropping it into small flashlight to running a drill to drill a hole in a piece of wood.
On the left we have Phil Mullins who was the mechanical designer for the whole hand. On the right is Pablo Garcia - the PM for the project.