Kowndinya Boyalakuntla

I’m a third-year Ph.D. student in Computer Science at Rutgers University, advised by Prof. Abdeslam Boularias.

I work on learning and planning for robot manipulation and whole-body control. My current research explores action abstractions from human motion capture for planning with learned world models, and Q-guided recovery for frozen imitation policies, toward reliable humanoid loco-manipulation.

Download CV Scholar GitHub Email

Ongoing research

Motion Abstractions for World-Model Planning

I am investigating how human motion capture can provide useful action abstractions for planning with learned world models. A learned motion representation maps the planner’s choice of motion and duration to a reference trajectory, which a closed-loop controller tracks. Starting with humanoid locomotion, I am studying when choosing these abstractions improves planning over fixed durations under matched time and compute budgets, with the longer-term goal of humanoid loco-manipulation.

Learning Recovery Proposals for Q-Guided Planning

I am exploring how to help a frozen imitation policy recover when it rarely proposes useful actions. The idea is to learn a small set of behaviorally diverse recovery proposals and use a shared task-value critic to guide their selection, keeping the proposals available at deployment. I plan to test whether reusable recovery proposals improve recovery and reduce repeated-search cost compared with Q-only selection, a single residual policy, and search without a learned proposal module. Comparisons will match interaction and compute budgets and measure whether performance on ordinary tasks is retained.

Publications

Google Scholar

PL-MPC architecture showing planning, replay, world-model and value learning, and actor learning.

PL-MPC: Beyond Policy Alignment: Closing the Planning–Learning Loop for Robot Control with Learned World Models

Kowndinya Boyalakuntla, Yuhan Liu, Abdeslam Boularias

Under review at ICRA 2027

Closing the loop between planning and learning for humanoid control and zero-shot sim-to-real manipulation.

Project PDF Code

TRACE architecture showing the frozen teacher plan, partial observations, student memory, and push or grasp decision.

TRACE: Plan-Conditioned Imitation for Robust Object Retrieval under Self-Occlusion in Dense Clutter

Kowndinya Boyalakuntla, Ajinkya Pawar, Abdeslam Boularias, Jingjin Yu

Under review at ICRA 2027

One initial plan, memory, and partial feedback guide target retrieval through dense clutter when the robot blocks its own view.

Project PDF Code

KARL: Kalman-Filter Assisted Reinforcement Learner for Dynamic Object Tracking and Grasping

Kowndinya Boyalakuntla, Abdeslam Boularias, Jingjin Yu

IROS, 2025

Tracking and grasping moving objects with an eye-in-hand camera, a Kalman filter, and reinforcement learning.

Paper Code

DAP: Diffusion-based Affordance Prediction for Multi-modality Storage

Haonan Chang, Kowndinya Boyalakuntla, Yuhan Liu, Xinyu Zhang, Liam Schramm, Abdeslam Boularias

IROS, 2024

Diffusion-based affordance prediction for placing objects in containers with different storage modes.

PDF Code

LGMCTS: Language-Guided Monte-Carlo Tree Search for Executable Semantic Object Rearrangement

Haonan Chang, Kai Gao, Kowndinya Boyalakuntla, Alex Lee, Baichuan Huang, Harish Udhaya Kumar, Jingjin Yu, Abdeslam Boularias

IROS, 2024

Turning language instructions into executable object rearrangements through geometric reasoning and tree search.

Project PDF Paper Code

Context-Aware Entity Grounding with Open-Vocabulary 3D Scene Graphs

Haonan Chang, Kowndinya Boyalakuntla, Shiyang Lu, Siwei Cai, Eric Pu Jing, Shreesh Keskar, Shijie Geng, Adeeb Abbas, Lifeng Zhou, Kostas Bekris, Abdeslam Boularias

CoRL, 2023

Grounding free-form language queries in open-vocabulary 3D scene graphs.

Project PDF Paper Code

Earlier researchSoftware engineering & document understanding

eGEN: an energy-saving modeling language and code generator for location-sensing of mobile apps

Kowndinya Boyalakuntla, Marimuthu Chinnakali, Sridhar Chimalakonda, Chandrasekaran Kalicheti

ESEC/FSE, 2022

A tool to specify energy-aware self-adaptive location-sensing strategies helping developers write battery-aware code for mobile apps.

Paper Code

RepoQuester: A Tool Towards Evaluating GitHub Projects

Kowndinya Boyalakuntla, Meiyappan Nagappan, Sridhar Chimalakonda, Nuthan Munaiah

ICSME, 2022

A command-line tool to assist developers in evaluating projects by providing quantitative scores for ten metrics that span project collaboration, quality and maintenance.

Paper Code