CROSS introduces a pre-commitment localization layer using continuous SE(3) pose branches and Gaussian-mixture filtering to reject false matches, improving long-term robot relocalization and semantic navigation under severe scene changes.
GeoWind2Plan predicts mission-time 3D urban wind via neural operators to enable energy-efficient UAV planning in seconds, reducing energy by up to 12.7% versus wind-agnostic paths.
SRL-MPC integrates reinforcement-learned parameter updates with shape-aware model predictive control via geometric separation features to navigate dense heterogeneous robot crowds safely and adaptively.
Differentiable nonlinear MPC computes parametric NLP sensitivities via IFT and smoothed IPM conditions within SQP, achieving over 3x speedups versus prior solvers.
Standard generative sequence models suffer physical misgeneralization, where local trajectory errors propagate through physical measurements to shift aggregate distributions; a data deviation kernel predicts these shifts and guides mitigation.
SOAR proposes regression-based LiDAR relocalization for UAVs using locality-preserving sliding-window attention and coordinate-independent initialization, achieving state-of-the-art accuracy on UAVLoc with a 40% higher success rate and over 10 meters lower mean error.
Inverse Learning trains forward/inverse models and hierarchical stacks for planning and control, matching offline RL and diffusion baselines on D4RL with far less compute while yielding smoother, near-optimal trajectories.
LCVN introduces a language-conditioned navigation benchmark and compares diffusion-based latent imagination against unified autoregressive prediction for embodied agents. Latent imagination yields more temporally coherent rollouts, while unified prediction generalizes better to unseen environments.