arXiv subjects

Rika Antonova

Publications and source records attributed to Rika Antonova.

At least 19 recordsLinked to original sources

arXiv2026

When Does Disaggregation Pay? Simulating Prefill--Decode--Attention--FFN Specialization for Agentic LLM Inference

Agentic inference now dominates the LLM inference landscape, requiring LLMs to actively engage in multi-turn interactions with tool-calling capabilities. This introduces a more complex workload for the underlying inference system: serving stages such as prefill and decode exhibit substantially different behaviors and demand distinct compute and memory-bandwidth capabilities. As a result, a single homogeneous GPU system now struggles to support agentic inference, motivating an industry shift toward heterogeneous systems with disaggregated serving capabilities, such as the emerging Vera-Rubin platform with GPUs and Groq LPUs. However, the question of what the optimal hardware should look like for each component in a heterogeneous system remains underexplored. To this end, we propose a novel simulation framework for disaggregated serving, termed \textbf{HeteroPanacea}, that enables system-level simulation across three dimensions: 1) disaggregated quantization, 2) automated intra- and inter-device parallelization scheduling, and 3) PDAF (prefill-decode-attention-FFN) NPU architectural heterogeneity. By combining these three axes, we provide a cross-stack simulation framework for future heterogeneous agentic serving systems. We confirm the benefit of Prefill Decode disaggregation, simulating increased serving throughput by up to 75\% compared to traditional serving with current GPUs and demonstrate 4 way Prefill Decode Attention FFN disaggregation is the most consistent for increasing throughput across different models, assuming custom NPUs. We also investigate the relationship between model architecture and gain from disaggregation by running a set of ablation studies.

cs.DC↗

arXiv2026

KernelCraft: Benchmarking for Agentic Close-to-Metal Kernel Generation on Emerging Hardware

New AI accelerators with novel instruction set architectures (ISAs) often require developers to manually craft low-level kernels, a time-consuming and error-prone process that does not scale across hardware targets. This delays emerging hardware platforms from reaching the market. While prior LLM-based code generation has shown promise in mature GPU ecosystems, it remains unclear whether agentic LLM systems can quickly produce valid and efficient kernels for emerging hardware with new ISAs. We present KernelCraft: the first benchmark for evaluating an LLM agent's ability to generate and optimize low-level kernels for customized accelerators through a function-calling, feedback-driven workflow. We evaluate agent performance across three emerging accelerators on more than 20 machine-learning tasks, each with five diverse task configurations. Across four leading reasoning models, the strongest agents generate functionally correct kernels for unseen ISAs within a few refinement steps and produce optimized kernels that match or outperform compiler baselines. These results demonstrate KernelCraft's potential to accelerate the accelerator chip development cycle. KernelCraft is available at https://kernelcraft-cam.github.io/.

cs.AR↗

arXiv2026

DOT-Sim: Differentiable Optical Tactile Simulation with Precise Real-to-Sim Physical Calibration

Simulating optical tactile sensors presents significant challenges due to their high deformability and intricate optical properties. To address these issues and enable a physically accurate simulation, we propose DOT-Sim: Differentiable Optical Tactile Simulation. Unlike prior simulators that rely on simplified models of deformable sensors, DOT-Sim accurately captures the physical behavior of soft sensors by modeling them as elastic materials using the Material Point Method (MPM). DOT-Sim enables rapid calibration of optical tactile sensor simulation using a small number of demonstrations within minutes, which is substantially faster than existing methods. Compared to current baselines, our approach supports much larger and non-linear deformations. To handle the optical aspect, we propose a novel approach to simulating optical responses by learning a residual image relative to the real-world idle state. We validate the physical and visual realism of our method through a series of zero-shot sim-to-real tasks. Our experiments show that DOT-Sim (1) accurately replicates the physical dynamics of a DenseTact optical tactile sensor in reality, (2) generates realistic optical outputs in contact-rich scenarios, (3) enables direct deployment of simulation-trained classifiers in the real world, achieving 85% classification accuracy on challenging objects and 90% accuracy in embedded tumor-type detection, and (4) allows precise trajectory following with a policy trained from demonstrations in simulation, with an average error of less than 0.9 mm.

cs.RO↗

arXiv2026

NPU Design for Diffusion Language Model Inference

Diffusion-based LLMs (dLLMs) fundamentally depart from traditional autoregressive (AR) LLM inference: they leverage bidirectional attention, block-wise KV cache refreshing, cross-step reuse, and a non-GEMM-centric sampling phase. These characteristics make current dLLMs incompatible with most existing NPUs, as their inference patterns, in particular the reduction-heavy, top-$k$-driven sampling stage, demand new ISA and memory hierarchy support beyond that of AR accelerators. In addition, the blocked diffusion KV cache breaks from the append-only paradigm assumed by AR NPUs, and conventional AR-derived KV quantization schemes were designed for static activation distributions and do not account for the step-wise distribution shifts introduced by iterative block-wise refinement in dLLMs. In this paper, we introduce the first NPU accelerator specifically designed for dLLMs. It delivers: a dLLM-oriented ISA and compiler; a hardware-optimized execution model for both the transformer inference and diffusion sampling used in dLLMs; a novel Block-Adaptive Online Smoothing (BAOS) for quantizing KV cache in dLLMs; and a complete RTL implementation synthesized in 7nm. To evaluate and validate our design, we introduce a tri-path simulation framework that comprises analytical, cycle-accurate, and accuracy simulators, together with cross-validations against physical hardware. The full NPU stack, including ISA, simulation tools, and quantization software, will be open-sourced upon acceptance.

cs.AR↗

arXiv2026

MemExplorer: Navigating the Heterogeneous Memory Design Space for Agentic Inference NPUs

Emerging agentic LLM workloads are driving rapidly growing demand on both memory capacity and bandwidth, with different phases of inference (e.g., prefill and decode) imposing distinct requirements. Industry is responding by composing heterogeneous accelerators into single interconnected systems, as exemplified by NVIDIA's Vera Rubin platform, where each device brings its own memory architecture. This heterogeneity is further compounded by a widening landscape of available memory technologies: high-density on-chip SRAM, HBM, LPDDR, GDDR, and emerging options such as high-bandwidth flash (HBF), each offering different capacity, bandwidth, and power trade-offs. Identifying the right memory architecture for next-generation inference accelerators requires navigating a vast and rapidly evolving design space, in which the interplay between workload characteristics, NPU design dimensions, and memory system design remains largely underexplored. To address this challenge, we present MemExplorer, a new memory system synthesizer for heterogeneous NPU systems. MemExplorer provides a unified abstraction for modeling diverse memory technologies across different hierarchy levels (e.g., on-chip and off-chip) and automatically determines an efficient heterogeneous memory system together with NPU design choices (e.g., matrix engine size) to balance throughput and power between prefilling and decoding devices in a multi-device NPU system. Experimental results show that, under the same power budget for agentic workloads, MemExplorer achieves up to 2.3x higher energy efficiency than the baseline NPU and 3.23x higher than H100 in the prefill-only setting. Under equivalent performance targets in the decode setting, it further delivers up to 1.93x and 2.72x higher power efficiency over the baseline NPU and H100, respectively.

cs.AR↗

arXiv2026

Combating the Memory Walls: Optimization Pathways for Long-Context Agentic LLM Inference

LLMs now form the backbone of AI agents across a diverse range of applications, including tool use, command-line interfaces, and web or computer interaction. These agentic LLM inference tasks are fundamentally different from chatbot-focused inference. They often involve much longer context lengths to capture complex and prolonged inputs, such as an entire webpage DOM or complicated tool-call trajectories. This, in turn, generates significant off-chip memory traffic during inference and causes workloads to be constrained by two memory walls, namely the bandwidth wall and the capacity wall, preventing compute units from achieving high utilization. In this paper, we introduce PLENA, a hardware-software co-designed system built around three core optimization pathways. PLENA features a novel flattened systolic-array architecture (Pathway 1) and efficient compute and memory units that support an asymmetric quantization scheme (Pathway 2). It also provides native support for FlashAttention (Pathway 3). In addition, PLENA includes a complete software-hardware stack, consisting of a custom ISA, a compiler, a transaction-level simulator, and an automated design-space exploration flow. Experimental results show that PLENA delivers up to 2.23x and 4.70x higher throughput than the A100 GPU and TPU v6e, respectively, under identical multiplier counts and memory configurations during LLaMA agentic inference. PLENA also achieves up to 4.04x higher energy efficiency than the A100 GPU. The full PLENA system, including its simulator, compiler, ISA, and RTL implementation, will be open-sourced to the research community.

cs.AR↗

arXiv2025

HoMeR: Learning In-the-Wild Mobile Manipulation via Hybrid Imitation and Whole-Body Control

We introduce HoMeR, an imitation learning framework for mobile manipulation that combines whole-body control with hybrid action modes that handle both long-range and fine-grained motion, enabling effective performance on realistic in-the-wild tasks. At its core is a fast, kinematics-based whole-body controller that maps desired end-effector poses to coordinated motion across the mobile base and arm. Within this reduced end-effector action space, HoMeR learns to switch between absolute pose predictions for long-range movement and relative pose predictions for fine-grained manipulation, offloading low-level coordination to the controller and focusing learning on task-level decisions. We deploy HoMeR on a holonomic mobile manipulator with a 7-DoF arm in a real home. We compare HoMeR to baselines without hybrid actions or whole-body control across 3 simulated and 3 real household tasks such as opening cabinets, sweeping trash, and rearranging pillows. Across tasks, HoMeR achieves an overall success rate of 79.17% using just 20 demonstrations per task, outperforming the next best baseline by 29.17 on average. HoMeR is also compatible with vision-language models and can leverage their internet-scale priors to better generalize to novel object appearances, layouts, and cluttered scenes. In summary, HoMeR moves beyond tabletop settings and demonstrates a scalable path toward sample-efficient, generalizable manipulation in everyday indoor spaces. Code, videos, and supplementary material are available at: http://homer-manip.github.io

cs.RO↗

arXiv2025

Group Policy Gradient

We introduce Group Policy Gradient (GPG), a family of critic-free policy-gradient estimators for general MDPs. Inspired by the success of GRPO's approach in Reinforcement Learning from Human Feedback (RLHF), GPG replaces a learned value function with a group-based Monte Carlo advantage estimator, removing the memory, compute, and hyperparameter costs of training a critic while preserving PPO's clipped-objective structure. We prove the consistency of the GPG estimator, analyze the bias-variance tradeoffs, and demonstrate empirically that GPG matches or outperforms PPO on standard benchmarks. GPG makes better use of parallel simulations, which, together with its critic-free design, results in more efficient use of computational resources than PPO.

cs.LG↗

arXiv2025

Mobi-$π$: Mobilizing Your Robot Learning Policy

Learned visuomotor policies are capable of performing increasingly complex manipulation tasks. However, most of these policies are trained on data collected from limited robot positions and camera viewpoints. This leads to poor generalization to novel robot positions, which limits the use of these policies on mobile platforms, especially for precise tasks like pressing buttons or turning faucets. In this work, we formulate the policy mobilization problem: find a mobile robot base pose in a novel environment that is in distribution with respect to a manipulation policy trained on a limited set of camera viewpoints. Compared to retraining the policy itself to be more robust to unseen robot base pose initializations, policy mobilization decouples navigation from manipulation and thus does not require additional demonstrations. Crucially, this problem formulation complements existing efforts to improve manipulation policy robustness to novel viewpoints and remains compatible with them. We propose a novel approach for policy mobilization that bridges navigation and manipulation by optimizing the robot's base pose to align with an in-distribution base pose for a learned policy. Our approach utilizes 3D Gaussian Splatting for novel view synthesis, a score function to evaluate pose suitability, and sampling-based optimization to identify optimal robot poses. To understand policy mobilization in more depth, we also introduce the Mobi-$π$ framework, which includes: (1) metrics that quantify the difficulty of mobilizing a given policy, (2) a suite of simulated mobile manipulation tasks based on RoboCasa to evaluate policy mobilization, and (3) visualization tools for analysis. In both our developed simulation task suite and the real world, we show that our approach outperforms baselines, demonstrating its effectiveness for policy mobilization.

cs.RO↗

arXiv2025

CUPID: Curating Data your Robot Loves with Influence Functions

In robot imitation learning, policy performance is tightly coupled with the quality and composition of the demonstration data. Yet, developing a precise understanding of how individual demonstrations contribute to downstream outcomes - such as closed-loop task success or failure - remains a persistent challenge. We propose CUPID, a robot data curation method based on a novel influence function-theoretic formulation for imitation learning policies. Given a set of evaluation rollouts, CUPID estimates the influence of each training demonstration on the policy's expected return. This enables ranking and selection of demonstrations according to their impact on the policy's closed-loop performance. We use CUPID to curate data by 1) filtering out training demonstrations that harm policy performance and 2) subselecting newly collected trajectories that will most improve the policy. Extensive simulated and hardware experiments show that our approach consistently identifies which data drives test-time performance. For example, training with less than 33% of curated data can yield state-of-the-art diffusion policies on the simulated RoboMimic benchmark, with similar gains observed in hardware. Furthermore, hardware experiments show that our method can identify robust strategies under distribution shift, isolate spurious correlations, and even enhance the post-training of generalist robot policies. Videos and code are made available at: https://cupid-curation.github.io.

cs.RO↗

arXiv2025

Causal-PIK: Causality-based Physical Reasoning with a Physics-Informed Kernel

Tasks that involve complex interactions between objects with unknown dynamics make planning before execution difficult. These tasks require agents to iteratively improve their actions after actively exploring causes and effects in the environment. For these type of tasks, we propose Causal-PIK, a method that leverages Bayesian optimization to reason about causal interactions via a Physics-Informed Kernel to help guide efficient search for the best next action. Experimental results on Virtual Tools and PHYRE physical reasoning benchmarks show that Causal-PIK outperforms state-of-the-art results, requiring fewer actions to reach the goal. We also compare Causal-PIK to human studies, including results from a new user study we conducted on the PHYRE benchmark. We find that Causal-PIK remains competitive on tasks that are very challenging, even for human problem-solvers.

cs.LG↗

arXiv2025

DiffCloud: Real-to-Sim from Point Clouds with Differentiable Simulation and Rendering of Deformable Objects

Research in manipulation of deformable objects is typically conducted on a limited range of scenarios, because handling each scenario on hardware takes significant effort. Realistic simulators with support for various types of deformations and interactions have the potential to speed up experimentation with novel tasks and algorithms. However, for highly deformable objects it is challenging to align the output of a simulator with the behavior of real objects. Manual tuning is not intuitive, hence automated methods are needed. We view this alignment problem as a joint perception-inference challenge and demonstrate how to use recent neural network architectures to successfully perform simulation parameter inference from real point clouds. We analyze the performance of various architectures, comparing their data and training requirements. Furthermore, we propose to leverage differentiable point cloud sampling and differentiable simulation to significantly reduce the time to achieve the alignment. We employ an efficient way to propagate gradients from point clouds to simulated meshes and further through to the physical simulation parameters, such as mass and stiffness. Experiments with highly deformable objects show that our method can achieve comparable or better alignment with real object behavior, while reducing the time needed to achieve this by more than an order of magnitude. Videos and supplementary material are available at https://diffcloud.github.io.

cs.RO↗

arXiv2025

Deformable Cargo Transport in Microgravity with Astrobee

We present pyastrobee: a simulation environment and control stack for Astrobee in Python, with an emphasis on cargo manipulation and transport tasks. We also demonstrate preliminary success from a sampling-based MPC controller, using reduced-order models of NASA's cargo transfer bag (CTB) to control a high-order deformable finite element model. Our code is open-source, fully documented, and available at https://danielpmorton.github.io/pyastrobee

cs.RO↗

arXiv2024

EquiBot: SIM(3)-Equivariant Diffusion Policy for Generalizable and Data Efficient Learning

Building effective imitation learning methods that enable robots to learn from limited data and still generalize across diverse real-world environments is a long-standing problem in robot learning. We propose Equibot, a robust, data-efficient, and generalizable approach for robot manipulation task learning. Our approach combines SIM(3)-equivariant neural network architectures with diffusion models. This ensures that our learned policies are invariant to changes in scale, rotation, and translation, enhancing their applicability to unseen environments while retaining the benefits of diffusion-based policy learning such as multi-modality and robustness. We show on a suite of 6 simulation tasks that our proposed method reduces the data requirements and improves generalization to novel scenarios. In the real world, with 10 variations of 6 mobile manipulation tasks, we show that our method can easily generalize to novel objects and scenes after learning from just 5 minutes of human demonstrations in each task.

cs.RO↗

arXiv2024

Unpacking Failure Modes of Generative Policies: Runtime Monitoring of Consistency and Progress

Robot behavior policies trained via imitation learning are prone to failure under conditions that deviate from their training data. Thus, algorithms that monitor learned policies at test time and provide early warnings of failure are necessary to facilitate scalable deployment. We propose Sentinel, a runtime monitoring framework that splits the detection of failures into two complementary categories: 1) Erratic failures, which we detect using statistical measures of temporal action consistency, and 2) task progression failures, where we use Vision Language Models (VLMs) to detect when the policy confidently and consistently takes actions that do not solve the task. Our approach has two key strengths. First, because learned policies exhibit diverse failure modes, combining complementary detectors leads to significantly higher accuracy at failure detection. Second, using a statistical temporal action consistency measure ensures that we quickly detect when multimodal, generative policies exhibit erratic behavior at negligible computational cost. In contrast, we only use VLMs to detect failure modes that are less time-sensitive. We demonstrate our approach in the context of diffusion policies trained on robotic mobile manipulation domains in both simulation and the real world. By unifying temporal consistency detection and VLM runtime monitoring, Sentinel detects 18% more failures than using either of the two detectors alone and significantly outperforms baselines, thus highlighting the importance of assigning specialized detectors to complementary categories of failure. Qualitative results are made available at https://sites.google.com/stanford.edu/sentinel.

cs.RO↗

arXiv2024

EquivAct: SIM(3)-Equivariant Visuomotor Policies beyond Rigid Object Manipulation

If a robot masters folding a kitchen towel, we would expect it to master folding a large beach towel. However, existing policy learning methods that rely on data augmentation still don't guarantee such generalization. Our insight is to add equivariance to both the visual object representation and policy architecture. We propose EquivAct which utilizes SIM(3)-equivariant network structures that guarantee generalization across all possible object translations, 3D rotations, and scales by construction. EquivAct is trained in two phases. We first pre-train a SIM(3)-equivariant visual representation on simulated scene point clouds. Then, we learn a SIM(3)-equivariant visuomotor policy using a small amount of source task demonstrations. We show that the learned policy directly transfers to objects that substantially differ from demonstrations in scale, position, and orientation. We evaluate our method in three manipulation tasks involving deformable and articulated objects, going beyond typical rigid object manipulation tasks considered in prior work. We conduct experiments both in simulation and in reality. For real robot experiments, our method uses 20 human demonstrations of a tabletop task and transfers zero-shot to a mobile manipulation task in a much larger setup. Experiments confirm that our contrastive pre-training procedure and equivariant architecture offer significant improvements over prior work. Project website: https://equivact.github.io

cs.RO↗

arXiv2023

TidyBot: Personalized Robot Assistance with Large Language Models

For a robot to personalize physical assistance effectively, it must learn user preferences that can be generally reapplied to future scenarios. In this work, we investigate personalization of household cleanup with robots that can tidy up rooms by picking up objects and putting them away. A key challenge is determining the proper place to put each object, as people's preferences can vary greatly depending on personal taste or cultural background. For instance, one person may prefer storing shirts in the drawer, while another may prefer them on the shelf. We aim to build systems that can learn such preferences from just a handful of examples via prior interactions with a particular person. We show that robots can combine language-based planning and perception with the few-shot summarization capabilities of large language models (LLMs) to infer generalized user preferences that are broadly applicable to future interactions. This approach enables fast adaptation and achieves 91.2% accuracy on unseen objects in our benchmark dataset. We also demonstrate our approach on a real-world mobile manipulator called TidyBot, which successfully puts away 85.0% of objects in real-world test scenarios.

cs.RO↗

arXiv2023

In-Hand Manipulation of Unknown Objects with Tactile Sensing for Insertion

In this paper, we present a method to manipulate unknown objects in-hand using tactile sensing without relying on a known object model. In many cases, vision-only approaches may not be feasible; for example, due to occlusion in cluttered spaces. We address this limitation by introducing a method to reorient unknown objects using tactile sensing. It incrementally builds a probabilistic estimate of the object shape and pose during task-driven manipulation. Our approach uses Bayesian optimization to balance exploration of the global object shape with efficient task completion. To demonstrate the effectiveness of our method, we apply it to a simulated Tactile-Enabled Roller Grasper, a gripper that rolls objects in hand while collecting tactile data. We evaluate our method on an insertion task with randomly generated objects and find that it reliably reorients objects while significantly reducing the exploration time.

cs.RO↗