This study evaluates six offline algorithms on robot manipulation tasks, highlighting data quality and model design importance.

Ajay Mandlekar, Danfei Xu, Josiah Wong et al.

DNN-based gesture recognition combined with potential field algorithms enables intuitive drone swarm trajectory and formation control, with an average error of 5.6cm.

Valerii Serpiva, Ekaterina Karmanova, Aleksey Fedoseev et al.

Proposed C2F-ARM algorithm uses coarse-to-fine Q-attention for efficient visual robotic manipulation, requiring only 3 demonstrations for training.

Stephen James, Kentaro Wada, Tristan Laidlow et al.

DiSECt: a fully differentiable FEM-based soft material cutting simulator with continuous contact and damage models, enabling efficient parameter calibration and control optimization.

Eric Heiden, Miles Macklin, Yashraj Narang et al.

Proposes two methods: failure diagnosis via parameter deviation search and synthetic experience correction, boosting robot grasp success rate from 40% to over 60%.

Alex Mitrevski, Paul G. Plöger, Gerhard Lakemeyer

Using sim-to-real RL with environment randomization, a bipedal robot learns proprioceptive stair traversal without external perception.

Jonah Siekmann, Kevin Green, John Warila et al.

FlingBot employs high-speed dynamic flinging for cloth unfolding, achieving over 80% coverage within 3 actions, outperforming static baselines.

Huy Ha, Shuran Song

Model-based driving policy using offline logs and Bellman equation evaluation, achieving 25% higher score with 40× less data.

Dian Chen, Vladlen Koltun, Philipp Krähenbühl

Introduces an actuator-aware kino-dynamic planner and hierarchical control enabling humanoid robot flips and spins in simulation.

Matthew Chignoli, Donghyun Kim, Elijah Stanger-Jones et al.

Tactile-RL method achieves insertion of unknown geometry objects with over 85% success rate.

Siyuan Dong, Devesh K. Jha, Diego Romeres et al.

LazyDAgger reduces context switches by 60% while improving success rates by 60% in robot learning tasks.

Ryan Hoque, Ashwin Balakrishna, Carl Putterman et al.

Proposed NDT-based radar odometry compatible with scanning and automotive radars, reducing translational errors by 51%/30%.

Pou-Chun Kung, Chieh-Chih Wang, Wen-Chieh Lin

Proposes a multi-stage multiple-shooting contact-implicit trajectory optimization (CIO) method with Newton impact law, achieving real-time dynamic object manipulation for 6-DOF robots.

Jean-Pierre Sleiman, Jan Carius, Ruben Grandia et al.

LbW framework combines unsupervised human-robot translation and keypoint detection, enabling single-video imitation learning with superior performance across five tasks.

Haoyu Xiong, Quanzhou Li, Yun-Chun Chen et al.

MP3 introduces an end-to-end mapless autonomous driving model using probabilistic scene representations for safe, interpretable decisions.

Sergio Casas, Abbas Sadat, Raquel Urtasun

Using GAIL for 6-DOF robot peg-in-hole tasks with 6μm precision.

Sagar Gubbi, Shishir Kolathaya, Bharadwaj Amrutur

Human-in-the-loop imitation learning via remote teleoperation improves robot task success in bottleneck regions.

Ajay Mandlekar, Danfei Xu, Roberto Martín-Martín et al.

Transformer-based topological planning with GNNs outperforms end-to-end VLN models, achieving 38.6% success rate on VLN-CE.

Kevin Chen, Junshen K. Chen, Jo Chuang et al.

Goal-conditioned Transporter networks with deep feature rearrangement enable flexible multi-step deformable object manipulation, achieving success rates up to 100% on simulated benchmarks.

Daniel Seita, Pete Florence, Jonathan Tompson et al.

FINO-Net fuses RGB, depth, and audio data to achieve 98.60% failure detection accuracy using deep multimodal neural networks.

Arda Inceoglu, Eren Erdal Aksoy, Abdullah Cihan Ak et al.