arXiv subjects

Divyansh Pandey

Publications and source records attributed to Divyansh Pandey.

6 recordsLinked to original sources

arXiv2026

Answering Without Arriving: An Evidence Audit for Embodied Question Answering

Embodied question answering asks an agent to navigate, observe, and then answer questions about what it saw, yet answer accuracy alone cannot tell whether any of this happened. A plausible answer can come from the question, the room label, or the answer choices, and a correct answer does not show that the supporting observation was ever made. We present a multi-room benchmark whose 200 test episodes and 800 questions are stratified by route extent, with an evaluation ladder that holds the questions fixed while varying only the evidence supplied to the answerer, from text-only probes through closed-loop agents to replay of the reference trajectory. Questions are generated from symbolic records of rooms, visible objects, and frames, so every answer can be traced to the observations that support it. A closed-loop vision-language agent reaches no target in 200 episodes yet answers 45.5% of questions, within 0.1 points of the same model given no images; replaying the reference trajectory raises the same model to 57.4%, so the evidence exists along the route but is not gathered. A family-level audit then shows that access to evidence is not the same as use of it. Four of eight families are largely answerable from text, one cites frames that cannot establish its own answer, and the oracle conditions reach 59.5% on Rooms-visited by counting the supplied images. Evidence acquisition, answer coverage, and shortcut sensitivity must therefore be established before an accuracy gain can be credited to memory.

cs.RO↗

arXiv2026

Do Embodied VLMs Actually Use Their Scene-Graph Memory?

Embodied agents that answer questions about their route need a memory of what they saw, and scene graphs are a natural form for it. A vision-language model (VLM) can receive such a graph as text or through a graph encoder, but accuracy alone cannot show whether the graph holds the episode's facts, reaches the model, or is actually relied on. In this diagnostic study, we introduce a graph-swap test for the last of these: each episode is given another episode's graph, and we count how many answers change. On an embodied question-answering benchmark with routes through two to seven rooms, we use it to compare ways of delivering scene-graph memory to a fine-tuned VLM. As text, the memory is used strongly, adding 30 points on route questions. As a graph, it is not, and we trace why: the encoder preserves the information, but standard pooled and late fusion discard it. Giving each node its own input token recovers part of it, a gain of 8.3 points on route questions that the swap test confirms comes from the graph. The test also exposes a shortcut that accuracy hides: when the same facts appear in the text, the model ignores its graph. The test thus offers a cheap check of whether a VLM uses its scene-graph memory, and our analysis points to how that memory should be delivered.

cs.RO↗

arXiv2026

Dissecting Transformers: A CLEAR Perspective towards Green AI

The rapid adoption of Large Language Models (LLMs) has raised significant environmental concerns. Unlike the one-time cost of training, LLM inference occurs continuously and dominates the AI energy footprint. Yet most sustainability studies report only coarse model-level metrics, treating energy efficiency as an afterthought rather than a primary objective. Addressing the limitation, we propose Component-Level Energy Assessment via Repetitions CLEAR, to overcome temporal mismatch between microsecond scale component execution and millisecond(ms) scale monitoring of energy sensors. Using CLEAR, we evaluate 15 models spanning four architecture types, keeping component-wise energy variance below 9.5% while capturing over 90% of total energy as individual components. We present the first comprehensive, fine-grained energy analysis of Transformer components across key parameters such as batch size, attention heads, hidden dimension, KV cache, and attention variants. Our findings reveal that Attention consumes significantly more Energy per FLOP as compared to the entire model, indicating that FLOPs alone fail to capture true component-level energy cost. CLEAR enables reliable fine-grained energy measurements and provides a strong formal foundation for predictive modelling of energy consumption.

cs.LG↗

arXiv2026

CALM: A Self-Adaptive Orchestration Approach for QoS-Aware Routing in Small Language Model based Systems

AI-enabled systems are subjected to various types of runtime uncertainties, ranging from dynamic workloads, resource requirements, model drift, etc. These uncertainties have a big impact on the overall Quality of Service (QoS). This is particularly true in the case of Language Model (LM) enabled systems where the autoregressive nature of token generation introduces variability in latency, energy usage and response quality. These systems, powered by LLMs, are either resource-intensive (if run on-prem) or raise privacy/cost concerns (if leveraged using APIs). While deploying a Small Language Model (SLM) can be resource-efficient, it often falls short in addressing the diversity and scale of real-world requirements. To this, we argue that, rather than relying on any one SLM, leveraging a coordinated fleet of SLMs, each with specialized strengths can enable systems to dynamically adapt to shifting contexts and workload patterns. However, realizing the full potential of such an approach demands intelligent orchestration and continuous adaptation. To this end, we introduce CALM , a self-adaptive orchestration mechanism based on MAPE-K. Our approach continuously monitors user queries, analyzes the QoS metrics of the SLMs, identifies the optimal SLM to be used, routes the query to the identified SLM and further to enhance the effectiveness and efficiency, leverages caching and scheduling to decide the SLMs to be kept in memory. Our evaluation shows that CALM reduces latency by approximately 40% and energy consumption by 50%, while preserving domain-specific task performance when compared to single-LLM baselines.

cs.SE↗

arXiv2025

POLARIS: Is Multi-Agentic Reasoning the Next Wave in Engineering Self-Adaptive Systems?

The growing scale, complexity, interconnectivity, and autonomy of modern software ecosystems introduce unprecedented uncertainty, challenging the foundations of traditional self-adaptation. Existing approaches, typically rule-driven controllers or isolated learning components, struggle to generalize to novel contexts or coordinate responses across distributed subsystems, leaving them ill-equipped for emergent unknown unknowns. Recent discussions on Self-Adaptation 2.0 emphasize an equal partnership between AI and adaptive systems, merging learning-driven intelligence with adaptive control for predictive and proactive behavior. Building on this foundation, we introduce POLARIS, a three-layer multi-agentic self-adaptation framework that advances beyond reactive adaptation. POLARIS integrates: (1) a low-latency Adapter layer for monitoring and safe execution, (2) a transparent Reasoning layer that generates and verifies plans using tool-aware, explainable agents, and (3) a Meta layer that records experiences and meta-learns improved adaptation policies over time. Through shared knowledge and predictive models, POLARIS handles uncertainty, learns from past actions, and evolves its strategies, enabling systems that anticipate change and maintain resilient, goal-directed behavior. Preliminary evaluation on two self-adaptive exemplars, SWIM and SWITCH, shows that POLARIS consistently outperforms state-of-the-art baselines. We argue this marks a shift toward Self-Adaptation 3.0, akin to Software 3.0: a paradigm where systems not only learn from their environment but also reason about and evolve their own adaptation processes, continuously improving to meet novel challenges.

cs.SE↗

arXiv2025

POSEIDON : Efficient Function Placement at the Edge using Deep Reinforcement Learning

Edge computing allows for reduced latency and operational costs compared to centralized cloud systems. In this context, serverless functions are emerging as a lightweight and effective paradigm for managing computational tasks on edge infrastructures. However, the placement of such functions in constrained edge nodes remains an open challenge. On one hand, it is key to minimize network delays and optimize resource consumption; on the other hand, decisions must be made in a timely manner due to the highly dynamic nature of edge environments. In this paper, we propose POSEIDON, a solution based on Deep Reinforcement Learning for the efficient placement of functions at the edge. POSEIDON leverages Proximal Policy Optimization (PPO) to place functions across a distributed network of nodes under highly dynamic workloads. A comprehensive empirical evaluation demonstrates that POSEIDON significantly reduces execution time, network delay, and resource consumption compared to state-of-the-art methods.

cs.DC↗