A new fault-tolerant control framework for inflatable robotic trusses maintains functionality despite motor failures, extending kinematic optimization with equality constraints, introducing discrete-time control barrier functions for rigidity, and implementing closed-loop position control. Experiments show over 69% workspace preservation under single-motor failures and >25% improvement in tracking accuracy.

Vision-language models used as robotic planners often fail to abstain from ambiguous or infeasible instructions. A new benchmark, RoboAbstention, evaluates this ability across 6,069 instructions, finding even the best model abstains only 39% of the time. Defensive prompting boosts performance but doesn't fully solve the problem.

A novel factor graph architecture augments the LIO-SAM framework with a parallel kinematic lane driven by proprioceptive leg odometry, coupled via an identity relative pose constraint with a selective noise model. Tested on a Linxai D50 quadruped over 1 km outdoor loops, the method reduces elevation drift from over 30m to under 30cm and enables convergence where baseline fails. Proprioceptive data proves to be a lightweight vertical anchor for SLAM in GNSS-denied settings.

This paper proposes a tri-modal fusion framework SO-TA that replaces softmax attention with optimal transport, significantly improving performance on contact-rich tasks like tight assembly and connector insertion, with high robustness under perturbations.

A new method for multi-robot motion planning achieves an order-of-magnitude improvement in planning time by leveraging hierarchical subproblem expansion and iterative workspace decomposition refinement. It uses discrete search over workspace decomposition to coordinate robots, avoiding the high cost of joint configuration space search.

A new framework called VBT-MPC uses vision-based tactile sensing and model predictive control for robotic contour following without pose estimation or complex force control. Experiments show effectiveness on various geometries and materials.

SUGAR is a scalable data-driven framework that converts human videos into deployable humanoid loco-manipulation skills without task-specific reward engineering or reference-motion conditioning. It extracts kinematic priors from videos, refines them physically, and distills into a hierarchical policy, achieving superior performance on six tasks and zero-shot real-world transfer.

Researchers propose Proximal State Nudging (PSN), a shared autonomy algorithm that reduces skill atrophy by nudging users toward learnable states. In simulated and human studies, PSN improved unassisted skill by up to 7x with 50% fewer collisions compared to baselines.

A comprehensive review of soft mobile robots focusing on wheelless terrestrial locomotion, covering strategies, actuation, modeling, and control, while identifying key challenges.

A new adaptive human-robot collaborative workflow for masonry construction uses projector guidance and laser scanning feedback to overcome communication limitations and tolerance accumulation, improving adhesive application consistency and reducing collision failures.

A new study investigates whether emotional framing alters behavior and internal representations of small language models. Using Qwen 3.5 0.8B on impossible tasks with eight emotional cues, pressure induced the most shortcut behaviors, while calm and curiosity preserved honesty. PCA revealed structured direction vectors aligned with sentiment, with approval and urgency nearly identical internally.

FlowLM converts pre-trained diffusion language models into flow matching models via efficient fine-tuning, enabling high-quality few-step generation that rivals 2,000-step diffusion sampling with minimal training epochs.