SenseAct™

SenseAct™ is a benchmark task suite for developing and evaluating reinforcement learning methods with physical robots by abstracting over the complexity of real-time control of robotic components. SenseAct’s guiding principles of minimizing delays and maximizing timing consistency via proactive computation lead to responsive learned behavior and reliable learning via state-of-the-art algorithms. All the task implementations share a core structure which can be reused to implement new robotic tasks that benefit from SenseAct’s tight control over system delays.

 

The implementation can be accessed from Github, which currently contains several tasks based on the following types of robots:

UR Robotic Arms

The UR series are collaborative industrial arms with six joints produced by Universal Robots.

 

UR-Reacher-2

Task: Move a UR arm end-effector to any given location on a two-dimensional plane by controlling only two joints. 

ur-reacher-6-trpo.gif

UR-Reacher-6

Task: Move a UR arm end-effector to any given location within a three-dimensional box space by controlling all six joints.

Dynamixels Servos

The Dynamixel (DXL) series are programmable Direct-Current servos manufactured by Robotis.

 
dxl-reacher-trpo.gif

DXL-Reacher

Task: Move the DXL servo joint to any given target position.

dxl-tracker-trpo.gif

DXL-Tracker

Task: Control the DXL servo joint to track a moving target.

Create 2 Mobile Bases

The Create 2 is a hobbyist version of iRobot’s Roomba vacuum robot.

 
create-mover-ppo.gif

Create-Mover

Task: Move the Create 2 forward as fast as possible within an enclosed arena by controlling its two wheels.

create-docker-trpo.gif

Create-Docker

Task: Dock the Create 2 in its charging station.

AI Research Blog