arXiv subjects

Aarav Shah

Publications and source records attributed to Aarav Shah.

7 recordsLinked to original sources

arXiv2026

Where Am I? Semantic Map Grounding via Vision-Language Models for Multi-Modal Localization

We address robot localization in GPS-denied indoor environments by reframing it as a semantic reasoning task rather than a geometric estimation problem. Motivated by how humans localize using object-level cues and labeled maps, we ask whether a vision-language model, given a front camera image, a polar LiDAR scan, and a top-down semantic grid map, can infer the robot pose. We fine-tune Qwen2.5-VL-7B with LoRA and attach a lightweight regression head that predicts continuous pose coordinates (x, y, theta) directly from the final hidden state, bypassing text generation. Training uses a composite position-and-direction loss with curriculum learning on a custom Gazebo dataset of 120,112 samples and 527 scenes. On the in-distribution test set of 18,017 samples, the model achieves 98.23 percent position accuracy, 98.00 percent direction accuracy, 96.75 percent full pose accuracy, a mean position error of 0.11 m, and a mean orientation error of 5.7 degrees at 0.62 s per sample. Position accuracy drops by only 7.2 percentage points on seven unseen object categories, reaching 90.99 percent, supporting semantic spatial reasoning rather than appearance memorization. With incomplete maps, fine-tuning recovers performance to 93.72 percent position accuracy, showing adaptability to stale or partial map information. Two ablations highlight cross-modal complementarity. Without LiDAR, using only camera and map inputs, position accuracy remains 95.06 percent, only 3.2 percentage points below the full system. However, when the camera sees no visible objects in a wall-facing view, LiDAR sustains 92.33 percent position accuracy, compared with 70.74 percent when neither LiDAR nor visible objects are available. This shows that LiDAR becomes the primary localization signal when camera semantics are unavailable and provides a reliable fallback under occlusion or sparse layouts.

cs.RO↗

arXiv2026

Non-Markovian Memory-Induced Effects in Quantum Cosmology

We study memory effects in quantum cosmology by extending the semiclassical Wheeler-DeWitt framework beyond its usual local form. The main idea is to introduce a causal memory kernel at sub leading order, rather than imposing fractional derivatives directly by hand. In this setting, fractional time evolution appears as an effective description of the underlying nonlocal dynamics. We apply the framework to cosmological perturbations in de Sitter space and find a correction to the primordial power spectrum with a characteristic $k^{3/4}$ scaling. This contribution mainly affects high $l$ CMB temperature anisotropies, in contrast with standard semiclassical quantum gravitational corrections, which are strongest at large angular scales. We also discuss how the same memory-dependent dynamics may affect primordial non-Gaussianity, producing scale dependent corrections to the bispectrum and possible deviations from the usual squeezed limit consistency relation. Since the memory coefficient controls short scale power, it may also influence structure formation and could require some tuning in order to give phenomenologically acceptable astrophysical environments. Finally, we suggest that a cyclic extension of the Hawking-Hartle no-boundary proposal may provide a setting in which the effective memory strength can evolve across successive cosmological histories. In this way, the framework gives a concrete realization of fractional quantum cosmology based on memory effects and also points to possible observational signatures of nonlocal quantum gravitational dynamics.

gr-qc↗

arXiv2026

Inflation and Primordial Perturbations in Fractal Cosmology

We study inflationary dynamics within the framework of fractal cosmology, where space is characterized by an effective non-integer dimension $D$. In our work, fractal effects are sourced through thermodynamic modifications at the cosmological horizon. Using the modified Friedmann and continuity equations, we then derive the modified slow roll parameter and their evolution for linear, cubic, Starobinsky ($R+R^2$) and Natural inflationary potentials, showing that the slow roll parameters get suppressed for $D<3$. We further derive a fractal extension of the Mukhanov-Sasaki equation by introducing an effective momentum $k_{\text{eff}}$, which captures the modification of spatial Laplacian due to fractality. This leads to explicit corrections to the scalar power spectrum and the spectral index $n_s$, depending on both $D$ and a fractional length scale $L$. Confrontation with Planck 2018 data constrains the effective dimension to a best-fit range of $2.7\lesssim D \lesssim 3$ for the Starobinsky model. Furthermore, in the case of Natural Inflation, fractal corrections relax the usual requirement of super-Planckian axion decay constants, opening a phenomenologically viable parameter space inaccessible in the standard $3+1$ dimensional cosmology.

gr-qc↗

arXiv2025

Particle production and Higgs reheating

Reheating is essential for transforming the cold, vacuum dominated Universe at the end of inflation into the hot thermal bath required by the Standard Model. In many well motivated inflationary models, however, the inflaton has no direct couplings to other fields, raising the question of how the Universe becomes repopulated with particles. We address this question within the framework of geometric reheating, where energy transfer occurs purely through gravitational effects. Focusing on a Higgs inflationary scenario with a non-minimal curvature coupling $\xi \phi^2 R$, we derive the post-inflationary dynamics and compute particle production using the Bogoliubov formalism. We show that the rapid, oscillatory evolution of the curvature scalar after inflaton acts as a time dependent gravitational pump, creating scalar spectator particles even in the absence of explicit interactions. This curvature driven production mechanism provides a natural and efficient route to reheating, demonstrating that gravity alone can initiate the standard thermal history and bridge inflation with radiation domination in minimal, coupling free models of the early Universe.

astro-ph.CO↗

arXiv2025

Cosmology in the De Donder-Weyl Formulation of Einstein-Cartan Gravity

We investigate torsion-driven cosmological dynamics within the framework of Einstein-Cartan gravity using the De Donder-Weyl Hamiltonian formalism, where the tetrad and Lorentz connection act as independent variables and the Hamiltonian includes quadratic Riemann Cartan corrections. Embedding this theory in an FLRW background, we derive the corresponding torsion-modified Friedmann equations and analyze their solutions across radiation and matter-dominated epochs. The commonly assumed power law form $a(t)=\beta t^{\alpha}$ is shown to generate multiple solution branches, many of which can be considered to be 'unphysical'. A hybrid solution, $a(t)=Ct^{\alpha}e^{Dt^{\beta}}$ emerges in the special case $g_1=0$, where the quadratic Riemann-Cartan term vanishes. For $g_1\not=0,$ the equations become nonlinear, precluding closed-form analytic solutions. These findings highlight the limitations of the power-law approximation and identify the restricted conditions under which torsion can coherently drive cosmic expansion.

gr-qc↗

arXiv2025

The Ultra Slow-Roll Phase Of Warm Inflation In Braneworld Cosmology

Slow-roll of the inflaton (inflationary field) defines the standard dynamics of the inflationary epoch. However, the inflaton deviates from slow-roll when it encounters an extremely flat region in the inflationary potential, and enters a phase dubbed Ultra Slow Roll (USR). In previous studies, there have been various theories which modify the theory of general relativity, all of them having different motivations based on different paradigms. Among these, braneworld gravity, motivated from string theory; is one of the most prominent theories as it provides a geometrical explanation for the weakness of gravity. In this article, we explore two possible braneworld background theories, the Randall-Sundrum (RS-II) model and the Dvali-Gabadadze-Porratai (DGP) model, and then realize an USR phase in a particularly interesting inflationary scenario, called warm inflation. In the warm inflationary scenario, a thermal radiation bath coexists with the inflationary energy density as an effect of the dissipative dynamics. We then derive inflationary slow roll parameters and the primordial power spectrum of scalar curvature perturbations in such a setup. We then numerically investigate the evolution of the inflaton and the primordial power spectrum of scalar curvature perturbations. Our analysis shows that the braneworld contributions become progressively suppressed as the USR conditions are made more stringent, indicating that the USR phase effectively diminishes brane-induced corrections to standard inflationary dynamics.

gr-qc↗

arXiv2025

PALADIN: Self-Correcting Language Model Agents to Cure Tool-Failure Cases

Tool-augmented language agents frequently fail in real-world deployment due to tool malfunctions--timeouts, API exceptions, or inconsistent outputs--triggering cascading reasoning errors and task abandonment. Existing agent training pipelines optimize only for success trajectories, failing to expose models to the tool failures that dominate real-world usage. We propose \textbf{PALADIN}, a generalizable framework for equipping language agents with robust failure recovery capabilities. PALADIN trains on 50,000+ recovery-annotated trajectories constructed via systematic failure injection and expert demonstrations on an enhanced ToolBench dataset. Training uses LoRA-based fine-tuning to retain base capabilities while injecting recovery competence. At inference, PALADIN detects execution-time errors and retrieves the most similar case from a curated bank of 55+ failure exemplars aligned with ToolScan's taxonomy, then executes the corresponding recovery action. This approach generalizes to novel failures beyond the training distribution, retaining 95.2\% recovery performance on unseen tool APIs. Evaluation across PaladinEval and ToolReflectEval demonstrates consistent improvements in Recovery Rate (RR), Task Success Rate (TSR), Catastrophic Success Rate (CSR), and Efficiency Score (ES). PALADIN improves RR from 32.76% to 89.68% (+57% relative) over ToolBench and outperforms the strongest baseline CRITIC (76.34%) by +13.3%. Against vanilla agents, PALADIN achieves 89.86\% RR (+66% relative improvement from 23.75%). These results establish PALADIN as an effective method for building fault-tolerant agents capable of robust recovery in real-world tool environments.

cs.LG↗