3D semantic scene-graph paper demonstrates open-vocabulary object retrieval on Spot - 2026-06-22 - Research Publication - The paper builds hierarchical 3D scene-graph forests using VLM and LLM reasoning over semantic relationships. The authors demonstrate downstream robotics use in an open-vocabulary object-retrieval task using a real-world… - source: arxiv.org
Trajectory-consistent flow-matching paper evaluates visuomotor policies on Spot - 2026-05-08 - Research Publication - The paper improves flow-matching policies for robust visuomotor policy learning by reducing compounding trajectory errors. Real-robot tasks include Boston Dynamics Spot and Franka FR3 systems. - source: arxiv.org
CART terrain-adaptation paper evaluates context-aware locomotion on Spot - 2026-04-15 - Research Publication - The CART paper introduces context-aware terrain adaptation using temporal sequence selection for legged robots. The authors evaluate the method in simulation with Unitree Go2 and ANYmal-C and in real-world experiments with… - source: arxiv.org
Morphology-conditioned world-model paper analyzes transfer from Spot to Unitree Go1 - 2026-04-09 - Research Publication - The paper proposes morphology conditioning for quadrupedal world models to reduce hardware lock-in. It analyzes how a world model trained on Boston Dynamics Spot fails on Unitree Go1 without embodiment-aware conditioning. - source: arxiv.org
Robot body-language paper studies legible navigation intent with Boston Dynamics Spot - 2026-04-03 - Research Publication - The paper compares expressive motion, lights, text, and audio as communication modalities for robot navigation intent. The online video study uses Boston Dynamics Spot scenarios to measure how people interpret upcoming… - source: arxiv.org
Pandora paper demonstrates articulated scene-graph use with Boston Dynamics Spot - 2026-03-30 - Research Publication - Pandora builds articulated 3D scene graphs from egocentric vision to enrich robot scene representations. The paper demonstrates a Boston Dynamics Spot application retrieving concealed target items using a 3D scene graph. - source: arxiv.org
Underground-mine navigation paper validates GPU-free autonomous stack on Spot - 2026-03-04 - Research Publication - The paper presents a fully autonomous underground-mine navigation stack for Boston Dynamics Spot running on a low-power Intel NUC without GPU or network dependency. The system combines LiDAR-inertial odometry… - source: arxiv.org
Hilbert-augmented RL paper validates multi-robot coverage trajectories on Spot - 2026-02-23 - Research Publication - The paper augments DQN and PPO with Hilbert space-filling priors for decentralized coverage and exploration. The approach is validated on a Boston Dynamics Spot legged robot executing generated trajectories in indoor… - source: arxiv.org
Thermal-radiation-field paper validates fire-aware navigation on Boston Dynamics Spot - 2026-02-22 - Research Publication - The paper builds real-time thermal radiation fields from registered depth and thermal images for mobile robot navigation near fire. The approach is validated on a Boston Dynamics Spot robot in controlled experimental… - source: arxiv.org
World-model inspection paper deploys anomaly detection on Boston Dynamics Spot - 2026-02-18 - Research Publication - The paper combines supervised failure classification and anomaly detection for autonomous inspection tasks. It uses inspection feeds from office and industrial sites and demonstrates real-time deployment on Boston Dynamics… - source: arxiv.org
ZEST paper transfers athletic skills to Boston Dynamics Atlas - 2026-02-01 - Research Publication - ZEST learns dynamic multi-contact humanoid skills from motion capture and video without contact labels or extensive reward shaping. The paper reports deployment on Boston Dynamics Atlas for behaviors including crawling… - source: arxiv.org
Spot terrain-classification paper uses proprioceptive sensing for traversability - 2025-08-22 - Research Publication - The paper develops a terrain classifier for Boston Dynamics Spot using proprioceptive signals from the robot instead of external perception. Field tests classify terrain types from leg and body signals such as joint… - source: arxiv.org
Spatial relationship dataset captures robot-acquired indoor scenes using Spot - 2025-06-14 - Research Publication - The paper introduces a spatial-relationship-aware robotics dataset of nearly 1,000 indoor images captured using Boston Dynamics Spot. The dataset labels object attributes, positions, and spatial relationships for… - source: arxiv.org
Spot RL paper deploys high-speed learned locomotion on hardware - 2025-04-24 - Research Publication - The paper describes high-performance reinforcement learning deployment on Boston Dynamics Spot using the Spot RL Researcher Development Kit. The authors optimize simulation parameters with distributional measures and… - source: arxiv.org
Learning More With Less: Sample Efficient Model-Based RL for Loco-Manipulation - 2025-01-17 - Research Publication - By combining the agility of legged locomotion with the capabilities of manipulation, loco-manipulation platforms have the potential to perform complex tasks in real-world applications. Spot is the hardware platform used… - source: arxiv.org
DTAA: A Detect, Track and Avoid Architecture for navigation in spaces with Multiple Velocity Objects - 2024-12-11 - Research Publication - Proactive collision avoidance measures are imperative in environments where humans and robots coexist. Moreover, the introduction of high quality legged robots into workplaces highlighted the crucial role of a robust… - source: arxiv.org
Zero-shot Object-Centric Instruction Following: Integrating Foundation Models with Traditional Navigation - 2024-11-12 - Research Publication - Large scale scenes such as multifloor homes can be robustly and efficiently mapped with a 3D graph of landmarks estimated jointly with robot poses in a factor graph, a technique commonly used in commercial robots such as… - source: arxiv.org
HARPER dataset paper studies human-pose forecasting from Boston Dynamics Spot - 2024-03-21 - Research Publication - HARPER is a dataset for 3D body pose estimation and forecasting in interactions between users and Boston Dynamics Spot. The dataset includes Spot built-in stereo-camera recordings synchronized with OptiTrack ground truth… - source: arxiv.org
Atlas rough-terrain paper demonstrates traversability-aware humanoid path planning - 2022-03-01 - Research Publication - The paper presents a humanoid navigation planner that uses traversability assessment over 2.5D height maps for rough terrain. The framework combines graph search, trajectory optimization, footstep planning, and balance… - source: arxiv.org
Autonomous Spot paper demonstrates long-range exploration in extreme environments - 2020-10-19 - Research Publication - The paper presents long-range autonomous exploration with Boston Dynamics Spot in extreme environments such as DARPA Subterranean Challenge settings. It integrates the NeBula autonomy architecture with legged mobility… - source: arxiv.org
Articles
Boston Dynamics begins production of Atlas humanoid - 2026-01-05 - The production-ready Atlas unveiled at CES moves the long-running humanoid programme toward industrial work, with 2026 fleets committed to Hyundai's RMAC and Google DeepMind.