robot-manipulation

InternVLA-A1.5 integrates pretrained vision-language models with future prediction in latent space to enable efficient robot manipulation with compositional generalization and long-horizon execution, achieving state-of-the-art results on simulation benchmarks.

Almond Robotics' Axol robot demonstrates folding a towel.

VLA-Corrector introduces a lightweight detect-and-correct inference framework that adaptively adjusts action horizons in Vision-Language-Action policies without retraining, improving robustness and efficiency in robot manipulation tasks.

Warp RL replaces additive residual corrections in reinforcement learning with an invertible, state-conditioned transformation of the base policy's action distribution using monotonic rational-quadratic spline flows, enabling adaptation of distribution shape, scale, and geometry under dynamics shifts. It matches or outperforms residual correction in ManiSkill3 manipulation tasks and achieves 30% faster task completion in a real robot peg-insertion task.

3D HAMSTER enhances robot manipulation by using a vision-language model with depth encoding to generate 3D trajectories for point cloud-based control, outperforming 2D-guided baselines.

This paper describes the prizewinning solution for the LeHome Challenge at ICRA 2026, where a two-armed robot learns to fold various garments using a novel RL approach with a self-contained value function, asynchronous training, and heavy sim-to-real augmentation.

The Geometric Action Model (GAM) repurposes a pretrained geometric foundation model (GFM) as a unified backbone for language-conditioned robot manipulation, achieving higher accuracy, robustness, and efficiency than existing foundation-model-scale baselines across simulation and real-world benchmarks.

Researchers propose APT, a two-stage training method that pretrains action experts on vision-action pairs before integrating language conditioning, significantly improving out-of-distribution instruction generalization for Vision-Language-Action policies.

AEGIS uses activation-probe early warning to switch to a stronger policy before failures compound in long-horizon robot manipulation, recovering twice as many failures as budget-matched escalation.

AHA-WAM is an asynchronous world-action model that uses dual Diffusion Transformers to decouple world prediction from action execution, achieving efficient long-horizon planning and real-time control. It achieves state-of-the-art performance on robotic manipulation tasks with up to 92.8% success on RoboTwin and 78.3% on real-world tasks, while reaching 24.17 Hz closed-loop control.

This paper introduces Geometric Primary Structure (GPS), a new representation for articulated parts perception in robot manipulation, enabling efficient VR-based annotation and achieving a 73% success rate without fine-tuning.

Light-WAM is a lightweight world action model for efficient robot manipulation that uses a compact video backbone and downsampled latent space for future-video supervision, achieving high performance with low inference latency.

VoLoAgent integrates vision-language models with robot capabilities for open-vocabulary long-horizon manipulation tasks, introducing a physical orchestrator that plans, monitors, and recovers using interruptible tools, and a benchmark called RoboVoLo for evaluation.

TBD-VLA introduces a discrete vision-language-action framework that combines block diffusion with autoregressive generation to achieve efficient temporal action modeling and faster inference, significantly outperforming prior VLA approaches in simulation and real-world manipulation tasks.

Dream.exe proposes an evaluation framework that uses robotic manipulation tasks to assess video generation models' understanding of physical reality, finding that visual quality does not predict executable motion accuracy.

AFUN proposes an affordance foundation model that predicts functional masks and 3D motion curves from RGB-D observations and language descriptions, enabling generalizable robot manipulation across diverse environments. The model outperforms baselines on multiple benchmarks and can be deployed for real-world tasks without fine-tuning.

RoboSemanticBench is a benchmark that diagnoses semantic grounding in action prediction for vision-language-action models, revealing that while robots can grasp objects, they fail to select semantically correct targets based on instruction semantics.

IntentVLA is a history-conditioned visual-language-action framework that improves robot imitation learning stability by encoding short-horizon intents from visual observations, addressing challenges from partial observability and ambiguous observations. It also introduces AliasBench, an ambiguity-aware benchmark for evaluating such methods.

RoboEvolve is a framework that co-evolves a VLM planner and VGM simulator for robotic manipulation, achieving data efficiency with only 500 unlabeled seed images and robust continual learning.

Google DeepMind introduces Gemini Robotics On-Device, an efficient VLA model optimized to run locally on robotic devices, enabling low-latency operation and offline capability while maintaining strong dexterous manipulation and task generalization. The model can be fine-tuned with as few as 50-100 demonstrations and comes with an SDK for developers.