Akhil Sathuluri

I am a PhD student at TU Munich, working on co-evolution of morphology and control of robotic systems at the Robot Systems Forschungsgruppe at LPL.

I previously worked as an Electrical Engineer in the Robotics and Automation team of Bajaj Auto Ltd., the largest three-wheeler and the third largest two-wheeler manufacturer in the world. I work on electrical sub-system design for special purpose robotic machines and edge based machine vision systems.

I graduated from the Department of Engineering Design, Indian Institute of Technology Madras. I was a member of the Robotics Lab, advised by Prof. Sandipan Bandyopadhyay, where I worked on the topics of design, analysis and control of parallel robots.

Email  /  GitHub  /  LinkedIn

profile photo



Research

project image

Extended-configuration-space modelling: mapping to reduced order models and real-time simulation of Lagrangian dynamics and control of parallel manipulators


Akhil Sathuluri
Advisor: Sandipan Bandyopadhyay
To be submitted, 2020
preprint / code /

This paper presents a method for the real-time simulation of Lagrangian formulation of the dynamics of parallel manipulators. A choice of generalised coordinates called the extended-configuration-space is introduced, which builds on the configuration, actuator and task-space coordinates. In particular, the proposed method presents the versatility to be transformed either into task-space or actuator- space models without sacrificing on the computational advantage. A comparative numerical study illustrates the accuracy and efficiency of the proposed method. Further, simulation studies on a semi-regular Stewart platform manipulator (SRSPM) and the 6-RSS manipulator are also presented.

project image

A Study on the Lagrangian Formulation of Dynamics with Applications to Control of Parallel Manipulators


Akhil Sathuluri
Advisor: Sandipan Bandyopadhyay
Dual Degree Report, 2019
report / code / slides /

Parallel manipulators are a class of robots characterised by their closed-loop architec- ture which gives them the advantages of high precision and load carrying capacity over their serial counterparts. Due to the presence of closed kinematic loops, the dynamics model consists of both actuated and passive joints related by sizeable symbolic expres- sions, evaluation of which hinders the speed of computer simulation. As a result, in practice, the dynamics model is often simplified, and control algorithms are used to compensate the model inaccuracies. Further, the computation of a model-based control input in realistic time scales with a precise formulation of the system remains a chal- lenge. Therefore, a dynamics formulation that would enable faster calculations without compromising on the fidelity of the model forms the central idea of the report.

project image

Root-tracking methods and their applications in simulations


Akhil Sathuluri
Advisor: Sandipan Bandyopadhyay
To be submitted, 2019
preprint / code /

This paper presents a comparative study of three different methods used for tracking a particular branch of solution amongst all the solutions arising from solving a set of non-linear equations. Given an initial estimate of the required root, keeping track of the branch it belongs to is a problem that commonly arises in simulating multi-body systems like a parallel manipulator or a cable-driven parallel robot (CDPR). In such cases, these roots represent feasible configurations of the manipulator. Hence, the accuracy and fast computation of the solutions are essential to ensure the safe movement of the manipulator. The primary objective of the paper is to highlight the implementation, present the comparison of three methods of tracking and discuss the context of their application. All the tracking methods are compared with the help of a simulation of the Stewart platform manipulator (SPM) following a desired path. Further, the implementation of such a method allowing a 3-3 CDPR to follow a given path is also demonstrated.

project image

MaMiC: Macro and Micro Curriculum for Robotic Reinforcement Learning


Manan Tomar, Akhil Sathuluri, Balaraman Ravindran
Proceedings of the 18th International Conference on Autonomous Agents and MultiAgent Systems, 2018
paper / arxiv /

Shaping in humans and animals has been shown to be a powerful tool for learning complex tasks as compared to learning in a randomized fashion. This makes the problem less complex and enables one to solve the easier sub task at hand first. Generating a curriculum for such guided learning involves subjecting the agent to easier goals first, and then gradually increasing their difficulty. This paper takes a similar direction and proposes a dual curriculum scheme for solving robotic manipulation tasks with sparse rewards, called MaMiC. It includes a macro curriculum scheme which divides the task into multiple subtasks followed by a micro curriculum scheme which enables the agent to learn between such discovered subtasks. We show how combining macro and micro curriculum strategies help in overcoming major exploratory constraints considered in robot manipulation tasks without having to engineer any complex rewards. We also illustrate the meaning of the individual curricula and how they can be used independently based on the task. The performance of such a dual curriculum scheme is analyzed on the Fetch environments.

project image

Sim-to-real transfer with industrial robots


Akhil Sathuluri, Manan Tomar
Industrial Research, 2018
code /

There has been significant research in solving manipulation tasks involving many research/collaborative robots. The objective of this work is to augment the capability of industrial robots by teaching them new tricks. The means of achieving this is by bridging the gap between industrial robots and state-of-the-art reinforcement learning-based controllers. Specifically, to learn controllers to solve non-prehensile manipulation tasks in pybullet and transfer them to the KUKA-KR5-Arc robot. The deployed hierarchical control architecture allows for the successful transfer of the simulated agent.

project image

MagNex — Expendable robotic surgical tooltip


Karthik Chandrasekaran, Akhil Sathuluri, Asokan Thondiyath
IEEE International Conference on Robotics and Automation (ICRA), 2017
paper /

This paper presents the design of a single use disposable compliant surgical tooltip for a tele-operated surgical robot. The proposed design aims at mitigating bio-fouling of surgical tools. By implementing a monolithic design for surgical tooltip, the cleaning and sterilization processes needed after every surgery for traditional surgical robotic tools is considerably simplified. The proposed design has 3 degrees of freedom (DOF) with a modified serpentine joint for enhanced buckling strength and off-axis stiffness. A magnetic force based coupling is proposed as a means of transferring power through a hermetic barrier to maneuver the tooltip. The tooltip is thus coupled through a hermetically sealed tool shaft preventing biological material from entering the tool shaft. By having a pluggable tooltip, modularity is also achieved by interchangeable tooltip instead of replacing the whole tool during surgery. A prototype of the proposed design is developed and its functional performance is validated. Owing to the magnetic nexus which forms the magnetic coupling, the tool is named MagNex.

Template and the modified template