Kartik Ramachandruni

I am a 4^th Year PhD student in Robotics (Interactive Computing department) at the Georgia Institute of Technology. I am advised by Prof. Sonia Chernova as part of the Robot Autonomy and Interactive Learning (RAIL) lab. My research focuses on enabling embodied AI agents to assist users with daily routines in unstructured and highly personalized human spaces with minimal task supervision.

I graduated from the Indian Institute of Technology Jodhpur with a B.Tech. in Mechanical Engineering, after which I worked as a robotics researcher for two years at the Tata Research and Innovation labs in Bangalore, India. I also did a summer research internship with the Google Cerebra team in New York during the summer of 2023, where I worked on developing uncertainty-aware LLM agents for UI automation tasks.

Email: kvr6 [at] gatech [dot] edu  |  Resume/CV  |  Scholar

I am currently working on user-assistive robots that learn user organizational preferences, such as object placement and arrangement preferences, from passive observation rather than explicit task instructions or demonstrations. My research aims to develop semantic reasoning techniques that can learn novel rearrangement preferences by integrating contextual cues from passive observations of partially arranged environments (e.g. a half-empty fridge or cabinet), with the aim of generalizing to previously unseen objects and households. My broader research interests include employing user interaction to resolve goal uncertainty in assistive tasks, user-adaptive task planning for human-robot collaboration, and robot imitation learning from video demonstrations.

Object rearrangement is the problem of enabling a robot to identify the correct object placement in a complex environment. Prior work on object rearrangement has explored a diverse set of techniques for following user instructions to achieve some desired goal state. Logical predicates, images of the goal scene, and natural language descriptions have all been used to instruct a robot in how to arrange objects. In this work, we argue that burdening the user with specifying goal scenes is not necessary in partially-arranged environments, such as common household settings. Instead, we show that contextual cues from partially arranged scenes (i.e., the placement of some number of pre-arranged objects in the environment) provide sufficient context to enable robots to perform object rearrangement \textit{without any explicit user goal specification}. We introduce ConSOR, a Context-aware Semantic Object Rearrangement framework that utilizes contextual cues from a partially arranged initial state of the environment to complete the arrangement of new objects, without explicit goal specification from the user. We demonstrate that ConSOR strongly outperforms two baselines in generalizing to novel object arrangements and unseen object categories. The code and data are available at https://github.com/kartikvrama/consor .

Collaborative human-robot task execution approaches require mutual adaptation, allowing both the human and robot partners to take active roles in action selection and role assignment to achieve a single shared goal. Prior works have utilized a leader-follower paradigm in which either agent must follow the actions specified by the other agent. We introduce the User-aware Hierarchical Task Planning (UHTP) framework, a communication-free human-robot collaborative approach for adaptive execution of multi-step tasks that moves beyond the leader-follower paradigm. Specifically, our approach enables the robot to observe the human, perform actions that support the human's decisions, and actively select actions that maximize the expected efficiency of the collaborative task. In turn, the human chooses actions based on their observation of the task and the robot, without being dictated by a scheduler or the robot. We evaluate UHTP both in simulation and in a human subjects experiment of a collaborative drill assembly task. Our results show that UHTP achieves more efficient task plans and shorter task completion times than non-adaptive baselines across a wide range of human behaviors, that interacting with a UHTP-controlled robot reduces the human's cognitive workload, and that humans prefer to work with our adaptive robot over a fixed-policy alternative.

Robots are increasingly transitioning from specialized, single-task machines to general-purpose systems that operate in diverse and dynamic environments. To address the challenges associated with operation in real-world domains, robots must effectively generalize knowledge, learn, and be transparent in their decision making. This survey examines Semantic Reasoning techniques for robotic systems, which enable robots to encode and use semantic knowledge, including concepts, facts, ideas, and beliefs about the world. Continually perceiving, understanding, and generalizing semantic knowledge allows a robot to identify the meaningful patterns shared between problems and environments, and therefore more effectively perform a wide range of real-world tasks. We identify the three common components that make up a computational Semantic Reasoning Framework: knowledge sources, computational frameworks, and world representations. We analyze the existing implementations and the key characteristics of these components, highlight the many interactions that occur between them, and examine their integration for solving robotic tasks related to five aspects of the world, including objects, spaces, agents, tasks, and actions. By analyzing the computational formulation and underlying mechanisms of existing methods, we provide a unified view of the wide range of semantic reasoning techniques and identify open areas for future research.

This paper proposes an end-to-end self-supervised feature representation network named Attentive Task-Net or AT-Net for video-based task imitation. The proposed AT-Net incorporates a novel multi-level spatial attention module to identify the intended task demonstrated by the expert. The neural connections in AT-Net ensure the relevant information in the demonstration is amplified and the irrelevant information is suppressed while learning task-specific feature embeddings. This is achieved by a weighted combination of multiple inter mediate feature maps of the input image at different stages of the CNN pipeline. The weights of the combination are given by the compatibility scores, predicted by the attention module for respective feature maps. The AT-Net is trained using a metric learning loss which aims to decrease the distance between the feature representations of concurrent frames from multiple view points and increase the distance between temporally consecutive frames. The AT-Net features are then used to formulate a reinforcement learning problem for task imitation. Through experiments on the publicly available Multi-view pouring dataset, it is demonstrated that the output of the attention module highlights the task-specific objects while suppressing the rest of the background. The efficacy of the proposed method is further validated by qualitative and quantitative comparison with a state-of-the-art technique along with intensive ablation studies. The proposed method is implemented to imitate a pouring task where an RL agent is learned with the AT-Net in Gazebo simulator. Our findings show that the AT-Net achieves 6.5% decrease in alignment error along with a reduction in the number of training iterations by almost 155k over the state-of-the-art while satisfactorily imitating the intended task.