It is well known that the choice of input features for a learning task can strongly impact the success of learning. Nobody had previously looked into an important input feature choice for robotic interception tasks, however: whether to represent the robot by its joint angle rotations or spatial link positions. In this work we considered, in a deep reinforcement learning context, a task in which a robot arm must use a stick attached to its end effector to intercept a projectile. While joint angles are a standard choice for robot representation, we found that link positions produced far better success in the interception task. We speculate that the neural network is better able to calculate important distances, such as the distance from the stick to the projectile, when given spatial link positions instead of joint angles. Code and trained policies are available on the project's LoboGit page.
John E. G. Baxter, Torin Adamson, Satomi Sugaya and Lydia Tapia, "Exploring Learning for Intercepting Projectiles with a Robot-Held Stick", IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2021.(pdf,video)
README
Software License
Model Data
Learned Policies Data
John Baxter, Mohammad R. Yousefi, Satomi Sugaya, Marco Morales and Lydia Tapia, "Deep Prediction of Swept Volume Geometries: Robots and Resolutions", IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2020.(pdf)
ReadMe(Download)
Robot | Network Script | Training Data |
---|---|---|
Baxter | Baxter7DOF | Baxter_14_22x22x21 |
Closed Loop (CL) | ClosedLoop9DOF | ClosedLoop_18_41x50x3 |
Kuka10 | Kuka7DOF_101010 | Kuka_14_10x10x10 |
Kuka15 | Kuka7DOF_151515 | Kuka_14_15x15x15 |
Kuka20 | Kuka7DOF_202020 | Kuka_14_20x20x20 |
Kuka25 | Kuka7DOF_252525 | Kuka_14_25x25x25 |
Kuka30 | Kuka7DOF_303030 | Kuka_14_30x30x30 |
Kuka80 | Unavailable* | Unavailable* |
Mico | Mico6DOF | Mico_12_15x15x15 |
Planar | Planar16DOF | Planar_32_1x10x100 |
Prismatic(PR) | Prismatic8DOF | Prismatic_16_88x88x2 |
Sawyer | Sawyer8DOF | Sawyer_16_26x26x26 |
UR3 | UR3_6DOF | UR3_12_15x15x15 |
UR5 | UR5_6DOF | UR5_12_20x20x20 |
UR10 | UR10_6DOF | UR10_12_30x30x30 |
* There was an issue with data conversion. Currently in progress.
The world has been gripped by the COVID-19 pandemic, and the rightful first reaction of many of us has been to protect the safety and health of ourselves and our families. However, life is not completely paused, and we still have responsibilities to our research, classes, and peers. Rapidly implemented workfrom- home protocols have resulted in a widely lamented lack of worklife balance and progress in research. This column tackles these woes by presenting advice collected from the Women in Engineering (WIE) Committee of the IEEE Robotics and Automation Society on their best practices for continuing work during this challenging time. It also shares some perspectives of robotics students on how they have been impacted by the pandemic.
Covid Lab Safety Signs[1]
[1]Elena Delgado and Lydia Tapia, "Robotics Research During a Pandemic", In IEEE Robotics & Automation Magazine (RAM), September 2020.(article)
What do flavorings, perfumes, hormones, neurotransmitters, and allergens have in common? They are all triggers of biological processes that start with a small molecule binding to a cell protein receptor. In our lab, we are interested in understanding the basic mechanisms of ligand-receptor binding. |