Waymo's Collision Avoidance Testing (CAT) framework evaluates SAE Level 4 ADS safety in urgent scenarios via virtual simulation and behavior models, showing a 15% collision reduction.

Kristofer D. Kusano, Kurt Beatty, Scott Schnelle et al.

Large-scale real-world evaluation shows modular navigation achieves 90% success, outperforming end-to-end methods in real environments.

Theophile Gervet, Soumith Chintala, Dhruv Batra et al.

Introduces MuJoCo MPC's Predictive Sampling as a simple baseline, competitive with advanced algorithms, enabling real-time behavior synthesis.

Taylor Howell, Nimrod Gileadi, Saran Tunyasuvunakool et al.

Proposes task-driven hybrid model reduction using linear complementarity systems, enabling real-time control with fewer modes; tested on synthetic and robotic tasks.

Wanxin Jin, Michael Posa

Optimization-based control using contact modeling and model simplification enables real-time dynamic legged robot motion.

Patrick M. Wensing, Michael Posa, Yue Hu et al.

DexPoint achieves cross-object sim-to-real transfer using point clouds and dexterous hands, with success rates over 80%.

Yuzhe Qin, Binghao Huang, Zhao-Heng Yin et al.

ROS 2 leverages DDS middleware for modular, real-time, secure robotics, enhancing industrial reliability.

Steve Macenski, Tully Foote, Brian Gerkey et al.

CFEAR-based radar odometry with motion compensation and multi-scan registration achieves 1.09% error at 5Hz, near lidar SLAM accuracy.

Daniel Adolfsson, Martin Magnusson, Anas Alhashimi et al.

PlanT employs object-level sparse features with Transformer, achieving expert-level driving scores on CARLA Longest6, 5.3× faster inference.

Katrin Renz, Kashyap Chitta, Otniel-Bogdan Mercea et al.

RING++ uses roto-translation invariant features for global localization on sparse LiDAR scans, ensuring global convergence without learning, outperforming state-of-the-art methods.

Xuecheng Xu, Sha Lu, Jun Wu et al.

VLMaps fuses pretrained visual-language features with 3D reconstruction for precise spatial open-vocabulary navigation.

Chenguang Huang, Oier Mees, Andy Zeng et al.

PTR uses CQL pre-training on the Bridge Dataset to learn unseen tasks on a real WidowX robot from only 10–15 demonstrations.

Aviral Kumar, Anikait Singh, Frederik Ebert et al.

DexGraspNet, a large-scale dataset of 1.32 million diverse, physics-validated dexterous grasps generated via a deep-accelerated optimization method for ShadowHand.

Ruicheng Wang, Jialiang Zhang, Jiayi Chen et al.

HULC++ combines visual affordance models with hierarchical language-conditioned policies, achieving 65.2% success on CALVIN with minimal annotations.

Oier Mees, Jessica Borja-Diaz, Wolfram Burgard

NLMap integrates CLIP/ViLD for open-vocabulary scene understanding, boosting robot long-horizon planning success to 61.8%.

Boyuan Chen, Fei Xia, Brian Ichter et al.

Proposes distributed guiding vector fields with virtual parameters for multi-robot path tracking and coordination, ensuring global convergence.

Weijia Yao, Hector Garcia de Marina, Zhiyong Sun et al.

RACER employs decentralized multi-UAV coordination via hgrid decomposition and CVRP path optimization for rapid exploration.

Boyu Zhou, Hao Xu, Shaojie Shen

This work transforms large language models (LLMs) into robot policy generators via hierarchical code synthesis, achieving 39.8% on HumanEval and enabling spatial reasoning and task generalization.

Jacky Liang, Wenlong Huang, Fei Xia et al.

Using model-free RL (SAC, DroQ) with regularization, a quadruped learns to walk in 20 minutes across diverse terrains, emphasizing system design and low-level control.

Laura Smith, Ilya Kostrikov, Sergey Levine

LATTE uses pre-trained language and vision models with Transformer to adapt 3D robot trajectories from natural language commands, no task prior needed.

Arthur Bucker, Luis Figueredo, Sami Haddadin et al.