Robotics Software Engineer @ Clutterbot

Building robots that
think, plan & act.

Robotics Software Engineer · Researcher

I work at the intersection of reinforcement learning, sim-to-real transfer and motion planning — turning manipulation research into robots that operate reliably in the messy real world.

📍 Bengaluru, India

View Projects → 📄 Résumé in LinkedIn
scroll
Priyansh Sinha
// about me

Robotics engineer with a researcher's mindset.

Hi, I'm Priyansh — a Robotics Software Engineer at Clutterbot, where I build RL pipelines, perception, and motion-planning systems for robots that manipulate objects in cluttered, real-world homes.

I completed my MS by Research in Computer Science (Robotics) at IIIT Hyderabad, focusing on contact-rich, non-prehensile mobile manipulation. My work spans asymmetric actor-critic RL, sim-to-real transfer, MoveIt-based planning, and point-cloud perception — from theory to production-grade ROS2 deployments.

2
Peer-reviewed publications
8.2
MS by Research CGPA
5+
Years in robotics
// toolbox

Skills & Stack

The tools I reach for across simulation, learning, planning and real hardware.

⚙️ Software & Tools

ROS2MoveItIsaac Sim Isaac LabMuJoCoPyTorch Scikit-learnOpen3DMATLAB DockerGit

💻 Languages

C++Python

🧠 Domains

Reinforcement LearningSim2Real Motion PlanningManipulation Computer Vision

🤖 Robotics Hardware

uFactory xArm (7-DOF)Mobile Robots Nvidia OrinIntel NUC Raspberry PiIntel RealSense KinectHTOF Depth 2D / 3D LiDAR
// career

Experience

From plant-floor automation to cutting-edge manipulation research and production robotics.

Feb 2025 — Present

Robotics Software Engineer

Clutterbot · Bengaluru, India
  • Designed RL pipelines using an asymmetric actor-critic architecture with depth-based perception, privileged simulation states and domain randomization for sim-to-real generalization.
  • Integrated MoveIt for motion planning & constraint validation, and optimized trajectory caching/retrieval to cut redundant planning cycles.
  • Built a modular point-cloud processing pipeline with geometric reasoning for robust manipulation in cluttered environments.
  • Hardened the stack with ROS2 lifecycle nodes and SOLID principles for scalable, production-ready deployment.
2024

Founding Team — Robotics

OpenDroid · USA · home-assistive mobile manipulator
  • Designed and built a home-assistive mobile manipulator end-to-end, including hardware procurement and ROS integration for navigation & manipulation.
  • Integrated an LLM for natural-language arm control and designed home-environment behaviours (e.g. "fetch an apple").
OpenDroid prototype OpenDroid prototype OpenDroid ROS integration
Jan 2022 — Dec 2024

Research Assistant

Robotics Research Center, IIIT Hyderabad · under Prof. Madhav Krishna & Dr. G. Nagamanikandan
  • Developed and tested autonomous navigation & motion-planning algorithms for an autonomous wheelchair and a mobile robot using ROS.
  • Teaching Assistant for the Mechatronics Designs and Systems course.
Jan 2024 — Aug 2024

Machine Learning Intern (remote)

xLM Continuous Validation · PA, United States
  • Integrated LLMs into web testing using agentic frameworks (CrewAI, LaVague).
  • Built a RAG model over the company's database for document generation using LangChain.
Jan 2023 — Jan 2024

Chief Robotics Head

ChargeKart · Hyderabad, India
  • Led a team to develop a high-payload mobile robot for charging E-vehicles in gated parking, including ROS & NAV2 integration.
Aug 2020 — Feb 2022

Graduate Engineer Trainee — Plant Automation

Hero MotoCorp · Gurgaon, India
  • Optimized a 7-DOF industrial robot program using TPM principles, reducing downtime by 10%.
  • Debugged PLC-based AGVs with Kaizen methodologies to minimize operational interruptions.
  • Integrated an AI vision camera (Five-Why + Kaizen) that eliminated barcode/part mismatches and cut error rates by 30%.
May 2019 — Jun 2020

Data Acquisition Engineer

Team Conrods — Official BAJA Team · Chennai, India
  • Integrated sensors onto the vehicle and built lap-data visualization for performance analysis.
// selected work

Projects

Research and engineering across manipulation, perception, planning and learning.

Mobile Manipulation — Whole-Body Control

Trajectory-optimization planners for non-prehensile manipulation (e.g. door opening) on a 10-DOF mobile manipulator, with a full navigation stack and low-level controllers. Watch the whole-body controller in action on the right.

ROSWhole-Body ControlOptimization
▶ Watch full demo
Residual learning setup

Residual Learning for Sim-to-Real

Neural-network residual learning to bridge the sim-to-real gap in manipulator striking actions, reducing control error. In collaboration with Dr. Samarth Brahmbhatt, Intel Labs.

PyTorchSim2RealManipulation
Visual odometry

Visual Odometry (Stereo + RGBD)

Indoor & outdoor visual odometry using stereo and RGBD cameras, integrating the LightGlue feature matcher into the VO pipeline for robust correspondence.

Computer VisionLightGlueSLAM
Trajectory optimization

Trajectory Optimization — Sampling MPC

Sampling-based Model Predictive Control for robot path planning, implementing the Model Predictive Path Integral (MPPI) update rule over a finite time horizon.

MPCMPPIPlanning
Imitation learning

Imitation Learning Ablation

Ablation study on a LeRobot arm comparing Action Chunking with Transformers (ACT), SmolVLA and Diffusion Policy; adapted the best performers onto a custom robotic platform.

ACTSmolVLADiffusion Policy
Friction estimation & segmentation

Perception & Estimation

Friction estimation of objects via an LSTM model to boost task accuracy, plus semantic image segmentation using a Gaussian Mixture Model + KNN for improved classification.

LSTMGMMKNN
Model-based controller experiment

Model-Based Controllers

Implementations of gravity-compensation, force-torque, hybrid force-position, null-space and operational-space controllers on manipulators.

Control TheoryForce ControlOperational Space
// machine learning

ML Projects

Classical machine-learning projects — built from the ground up to understand the algorithms, not just call them.

01
🐦

Viral Tweet Prediction

Predicting whether a tweet goes viral from engagement and content features.

K-Nearest NeighborsClassification
02
✉️

Spam SMS Classification

Filtering spam from ham messages using a probabilistic bag-of-words model.

Multinomial Naive BayesNLP
03
🧑‍💻

Facial Recognition System

Eigenfaces-based face recognition via dimensionality reduction on image data.

PCAComputer Vision
04
🔁

Unsupervised Domain Adaptation

Aligning source and target domains using gradient-reversal back-propagation.

Back-propagationDomain Adaptation
// research

Publications & Thesis

Advances in Robotics (AIR) 2025 · IIT Jodhpur July 2025

Targeted Object Striking for a 7-DoF Manipulator: A Residual Learning Approach

Priyansh Sinha, Rishin Chakroborty, Dr. Samarth Brahmbhatt, Dr. G. Nagamanikandan

🌐 Project page →
IEEE CASE 2024 · Bari, Italy August 2024

A Hierarchical Manipulation Planning Framework Combining Striking, Pushing, and Pick & Place Motion Primitives

Priyansh Sinha, Prakrut Kotecha, Dr. G. Nagamanikandan

📄 IEEE Xplore → ▶ Experiment video →
MS Thesis · IIIT Hyderabad

Planning and Control Strategies for Contact-Rich, Non-Prehensile Mobile Manipulation

// background

Education & Coursework

MS by Research, Computer Science (Robotics)
IIIT Hyderabad, India
Jan 2023 — Dec 2024 · 8.2 CGPA
B.Tech, Mechatronics Engineering
SRM Institute of Science & Technology, Chennai
2016 — 2020 · 78%
Statistical Methods in AI — K-NN, Naive Bayes, GMM, Regression, MLE, MLP, CNN, RNN, LSTM, GAN
Computer Vision — RANSAC, Triangulation, Stereo Rectification, Optical Flow, Graph Cuts, GrabCut, CNN, GAN, Transformers
Robotics — Kinematics, Dynamics, Controls, Planning & Navigation (ICP, SLAM)

Let's build the next generation of robots.

I'm always happy to talk robotics, research collaborations, or interesting manipulation problems. Reach out anytime.

✉ sinhapriyansh.15@gmail.com in LinkedIn 📄 Résumé