Sudharshan Suresh | Suddhu

suddhu [at] cmu [dot] edu

I'm a PhD candidate in the Robotics Institute at Carnegie Mellon University, where I work with Michael Kaess in the Robot Perception Lab. I work on statistical methods for mapping and state-estimation, specifically towards robotic tactile perception. I'm also a visiting researcher at Meta AI, where I work with Mustafa Mukadam.

I completed my Masters in Robotics at CMU working with Michael Kaess on visual localization and active exploration for AUVs (thesis). Prior to that, I worked with Red Whittaker on state-estimation for lunar rovers, and spent time at IISc Bangalore working on visual understanding. In my undergrad, I majored in Controls and Instrumentation at NIT Trichy.

CV  /  Scholar  /  Github  /  LinkedIn


[Oct '22]  

Successfully passed my Ph.D. thesis proposal!

[Sep '22]  

MidasTouch was accepted to CoRL 2022 as an oral.

[Aug '22]  

We've extended iSDF for neural mapping with the Franka robot, code here.

[May '22]    

Organized the Debates on the Future of Robotics Research workshop at ICRA '22

[April '22]    

Spending the summer at Meta AI with Mustafa Mukadam working on tactile perception/SLAM!

[Jan '22]    

ShapeMap 3-D was accepted to ICRA 2022, with an open-source implementation.



MidasTouch: Monte-Carlo inference over distributions across sliding touch
S. Suresh, Z. Si, S. Anderson, M. Kaess, and M. Mukadam
Proc. Conf. on Robot Learning, CoRL, Dec 2022 (to appear)
[Oral: 6% acceptance rate]
paper / website / code / presentation
Tracking the pose distribution of a robot finger on an object surface over time, using surface geometry captured by a tactile sensor
ShapeMap 3-D: Efficient shape mapping through dense touch and vision
S. Suresh, Z. Si, J. Mangelson, W. Yuan, and M. Kaess
IEEE Intl. Conf. on Robotics and Automation, ICRA, May 2022
paper / website / code / presentation
Can we efficiently reconstruct household objects with touch and vision? We harness the GelSight sensor and a depth-camera for 3-D shape perception, as inference on a spatial graph informed by a Gaussian process.
Tactile SLAM: Real-time inference of shape and pose from planar pushing
S. Suresh, M. Bauza, K.-T. Yu, J. Mangelson, A. Rodriguez, and M. Kaess
IEEE Intl. Conf. on Robotics and Automation, ICRA, May 2021
[Finalist for the 2021 IEEE ICRA Best Paper Award in Service Robotics]
paper / website / presentation
Can we estimate object shape and pose in real-time through purely tactile sensing? We demonstrate this for planar pushing, combining Gaussian process implicit surfaces with factor-graph based inference.
Active SLAM using 3D submap saliency for underwater volumetric exploration
S. Suresh, P. Sodhi, J. Mangelson, D. Wettergreen, and M. Kaess
IEEE Intl. Conf. on Robotics and Automation, ICRA, May 2020
paper / presentation
How do you balance volumetric exploration and pose uncertainty in exploration? We combine a sampling-based planner, deformable pose graph, and a 3D saliency metric to explore a 3D underwater volume.
Through-water stereo SLAM with refraction correction for AUV localization
S. Suresh, E. Westman, and M. Kaess
IEEE Robotics and Automation Letters (RA-L), presented at ICRA 2019, Jan 2019
paper / presentation
How can you incorporate refraction into water-to-air visual SLAM? We present a novel method inspired by multimedia photogrammetry for underwater localization.
Localized imaging and mapping for underwater fuel storage basins
J. Hsiung, A. Tallaksen, L. Papincak, S. Suresh, H. Jones, W. L. Whittaker, and M. Kaess
Proceedings of the Symposium on Waste Management, Phoenix, Arizona, Mar 2018
paper / slides / video
What's the ideal sensor suite for underwater dense mapping? We build and demonstrate an inspection solution comprising of a stereo camera, IMU, standard + structured lighting, and depth sensor.
Camera-Only Kinematics for Small Lunar Rovers
S. Suresh , E. Fang, and W. L. Whittaker
Robotics Institute Summer Scholars Working Paper Journal, Nov 2016
Annual Meeting of the Lunar Exploration Analysis Group, Nov 2016
paper / video / poster
Is it possible to track a lunar rover's kinematic state through self-perception? With a downward-facing fisheye lens, we estimate the Autokrawler's kinematics on rugged terrain.
Object category understanding via eye fixations on freehand sketches
R. K. Sarvadevabhatla, S. Suresh and R. V. Babu
IEEE Transactions on Image Processing (TIP), May 2017
paper / website / dataset
Can we better understand free-hand sketches through human gaze fixations? We collect the SketchFix-160 dataset and investigate visual saliency to reveal multi-level consistency in sketches.


Other projects

Franka iSDF: neural mapping for tabletop scenes
S. Suresh, J. Ortiz, and M. Mukadam
Extending iSDF to build real-time neural models of tabletop scenes with the Franka Panda arm
DeepGeo: photo localization with deep neural network
S. Suresh, N. Chodosh, and M. Abello
arXiv / github
A deep network that beats humans at GeoGuessr, trained on our 50States10K dataset.
Task and motion planning for robotic food preparation
S. Suresh, T. Rhodes, M. Abello, and H. Yadav
pdf / video 1 / video 2
Hierarchical task and motion planning for a 6-DOF robot arm, to prepare yogurt parfaits!
Thin structure reconstruction via 3D lines and points
S. Suresh and M. Abello
Reconstructing thin objects in a scene through an SfM pipeline can be hard!
Factor graph optimization for dynamic parameter estimation
S. Suresh, E. Dexheimer, and M. Abello
We implement a method for estimation of MAV poses and dynamic parameters during flight.

Last updated: Oct 2022

Imitation is the highest form of flattery