Rohan P. Singh

Rohan P. Singh

twitter / github / scholar / linkedin / lab
email: rohan565singh[at]gmail.com

[pin] Times Square, NYC
ca. 2017

Hi,

I am a postdoctoral researcher at the CNRS-AIST Joint Robotics Lab (JRL) in Tsukuba, Japan. I received the Ph.D. degree from the University of Tsukuba in April 2024 researching on Reinforcement Learning for humanoid locomotion and the MS degree in April 2021, working on 6-DoF object pose estimation for grasping and manipulation.

I have been working at JRL since October, 2017, first as a full-time robotics engineer and then as an Research Assistant (RA) since April, 2019. All this time, I have been extremely fortunate to be mentored and advised by Fumio Kanehiro sensei.

If you are an acquaintance, a friend, a former coworker, or know me even remotely, I'm open to your (anonymous) feedback.

Research

My Ph.D. research focussed on developing reinforcement learning based controllers for practical applications of real humanoid robots. This is challenging in several ways, some of which are due to the large size and high joint friction of robots like the HRP-5P.

For my master's, I worked on robotic perception and grasping - using CNNs and stacked-hourglass networks for 6-DoF pose estimation, with minimal human effort involved in data labelling and 3D modelling.

I'm also proud to have worked at UAS-DTU during my undergrad in New Delhi. It's a fun, yet an exceptionally passionate and hardworking group of student engineers. Go check them out!

Upcoming

[NEW] Humanoid Locomotion on Compliant and Uneven Terrain with Deep RL

Reinforcement Learning policy on the HRP-5P humanoid.

News

6 Dec, 2024	I gave a talk at Tokyo AI - AI in Hardware and Robotics on "How to train your humanoid".
27 Nov, 2024	I gave a talk at the Dynamic Legged Systems (DLS) lab at IIT Genoa, Italy about my postdoc research on multi-contact humanoid locomotion, and sim-to-real methods.
26 Nov, 2024	I visited the Robotic Systems Lab (RSL) at ETH Zurich, Switzerland.
25 Nov, 2024	I gave a talk at the Computational Robotics Lab (CRL) lab at ETH Zurich, Switzerland about my past research on bipedal locomotion using reinforcement learning.
12 Nov, 2024	I attended the Trilateral AI Conference 2024. Generative AI: Pathways to Democratization, Transparency and Sustainability.
16 May, 2024	I presented my research at the ICRA 2024: Late Breaking Results Poster Session!
1 Apr, 2024	I started as a Postdoctoral researcher at CNRS-AIST JRL under Mitsuharu Morisawa!
5 Mar, 2024	I gave a talk (remotely) at the Gepetto group, LAAS-CNRS, France.
5 Feb, 2024	Today, I successfully defended my Ph.D. thesis! Thanks to the jury members: Olivier Stasse, Eiichi Yoshida, Jun Izawa, Ryusuke Sagawa.
5 Oct, 2023	I gave a talk at the humanoids team in Toyota Motor Corporation, Future Creation Center, Suidobashi, Tokyo
1 Nov, 2021	I was awarded the prestigious and generous JST SPRING Fellowship! The fellowship period is 2021/10/01 ~ 2024/03/31!
16 Feb, 2021	I was awarded the Excellent Master's Thesis Award (優秀修士論文賞)!

Publications

(Please see my Google Scholar for the full list.)

	Learning bipedal walking for humanoids with current feedback Rohan P. Singh, Zhaoming Xie, Pierre Gergondet, Fumio Kanehiro IEEE Access 2023
	mc-mujoco: Simulating Articulated Robots with FSM Controllers in MuJoCo Rohan P. Singh, Pierre Gergondet, Fumio Kanehiro IEEE/SICE International Symposium on System Integration (SII), 2023, USA Jan. 2023
	Learning Bipedal Walking On Planned Footsteps For Humanoid Robots Rohan P. Singh, Mehdi Benallegue, Mitsuharu Morisawa, Rafael Cisneros, Fumio Kanehiro IEEE-RAS 21st International Conference on Humanoid Robots (Humanoids), 2022, Japan Dec. 2022
	Rapid Pose Label Generation through Sparse Representation of Unknown Objects Rohan P. Singh, Mehdi Benallegue, Yusuke Yoshiyasu, Fumio Kanehiro International Conference on Robotics and Automation (ICRA), 2021, China (virtual) Jun. 2021
	Instance-specific 6-dof object pose estimation from minimal annotations Rohan P. Singh, Iori Kumagai, Antonio Gabas, Mehdi Benallegue, Yusuke Yoshiyasu, Fumio Kanehiro IEEE/SICE International Symposium on System Integration (SII), 2020, USA Feb. 2020
	Easy Pose Annotation of Real RGB-D Data for Known Objects Rohan P. Singh, Mehdi Benallegue, Yusuke Yoshiyasu, Fumio Kanehiro The Proceedings of JSME annual Conference on Robotics and Mechatronics (ROBOMECH) 2020, Japan (virtual) May 2020
	APE: A More Practical Approach To 6-Dof Pose Estimation Antonio Gabas, Yusuke Yoshiyasu, Rohan Pratap Singh, Ryusuke Sagawa, Eiichi Yoshida IEEE International Conference on Image Processing (ICIP), 2020, UAE (virtual) Oct. 2020

Projects

LearningHumanoidWalking
Training a humanoid robot for locomotion using Reinforcement Learning.
MuJoCo code for training the JVRC-1 humanoid robot with a Ray-based implementation of PPO.

mc-mujoco
Interface for mc-rtc with MuJoCo.
This makes it very easy to simulate any robot with any controller as long as they can be expressed in mc-rtc.

MuJoCo XML for JVRC1 / mujoco-python-viewer
Open-source model files for the JVRC1 Humanoid Robot.
This is a more realistic humanoid robot model than the one in OpenAI Gym.
Show more...

Install MuJoCo viewer, clone Git repository:

 $ pip install mujoco-python-viewer $ git clone https://github.com/isri-aist/jvrc_mj_description

Then, in Python:

 import mujoco import mujoco_viewer model = mujoco.MjModel.from_xml_path('xml/jvrc1.xml') data = mujoco.MjData(model) # create the viewer object viewer = mujoco_viewer.MujocoViewer(model, data) # simulate and render for _ in range(100000): mujoco.mj_step(model, data) viewer.render() # close viewer.close()

ANA Avatar XPRIZE
I'm a member of Team Janus
We are building teleoperated human-like Avatar robots!

RapidPoseLabels
Easily generate labeled data for 6-DoF Object Pose Estimation, even if you don't have a model
Work published at ICRA, 2021. Contact me if you want help on how to use this!

ObjectKeypointTrainer
Train CNNs for Object Keypoint Detection (6-DoF Pose Estimation)
Contact me if you want help on how to use this!