Robotics combines mechanics, electronics, and AI to build machines that can act in the physical world. Boston Dynamics and its Atlas robot are a concrete example of advanced locomotion driven by machine learning.
AURA-Mem introduces a constant-size recurrent memory (4,224 bytes) for robot policies, with a learned gate that writes only when observations would change the next action. On LIBERO-Long with OpenVLA-OFT 7B, it matches baseline policy (0.233 success) while reducing memory writes by 7× and VRAM by 6,061× versus KV-cache.
Study of three approaches to automatically generate enemy morphologies in video games based on collision data. The tested methods match or outperform an evolutionary baseline adapted from robotics morphology work.
Helsing, a German startup, unveils the RX-1 on June 1, 2026. The European military robot relies on American inference silicon, exposing Europe's technological dependence in defense robotics despite sovereignty ambitions.
Researchers from the University of Florida developed BlueME, an antenna enabling robot communication at 700 meters underwater depth. The technology addresses a major challenge in deep-sea robotic operations.
Mecka AI raises $60 million to accelerate AI robot development. The funding supports training of AI-based robotic models.
Nvidia unveils Cosmos 3, an AI model presented at Taipei GTC alongside humanoid robot Isaac GROOT. The model aims to enhance understanding of real-world scenes.
NVIDIA releases Cosmos 3, a collection of omnimodal world models (Nano 16B, Super 64B) capable of generating dynamic video, image, audio, and action commands from text, image, video, and action trajectory inputs. Available on Hugging Face for Physical AI applications.
Nvidia unveils Isaac GR00T, a humanoid robot presented by Jensen Huang. The reference model combines manipulation capabilities and learning features.
CTRL-STEER introduces a closed-loop control framework for Vision-Language-Action (VLA) models. Instead of fixed steering coefficients, it adaptively adjusts intervention strength over time using PID or reinforcement learning controllers. Experiments on OpenVLA with LIBERO task suites demonstrate improved concept regulation stability and better steering-task success trade-offs without retraining the base model.
Demo2Reward optimizes VLM reward model language instructions at test-time using 3-10 expert demonstrations to reduce false positives in robotics. No additional training required. Validated on simulated tasks and real-world transfer.
Parameterized Diffusion Policy (PDP) conditions diffusion policies on continuous parameters embedded in a learned behavior manifold. The approach enables smooth interpolation between strategies and efficient adaptation to novel constraints without weight updates. Improved performance on multimodal benchmarks in simulated and real-robot experiments.
Developer trains a local model to detect mosquitoes and eliminate them with a laser. Applied robotics project combining vision and automated control.
Nvidia launches three physical AI models at GTC Taipei: Cosmos 3 (world model), Alpamayo 2 Super (scaled-up autonomous driving model), and an open reference platform for humanoid robots.
Nvidia announces Alpamayo 2 Super 32B, positioned as the foundation of its robotaxi stack. Embedded performance metrics remain unpublished, raising operational questions for automotive R&D directors.
OpenAI restarts robotics team five years after shutdown, emerging from world simulation research program. Near-term goal: infrastructure robots. Long-term vision from CEO Sam Altman: personal robot for everyone.
BYD launches its Xuanji A3 chip for autonomous driving, reducing reliance on Nvidia. The Chinese automaker aims to control its value chain by developing proprietary autonomous driving solutions.
NVIDIA releases Cosmos 3, an open omni-model for physical AI that reasons and acts. The model processes video, text, and images to understand real-world physics and generate robotic actions.
Reinforcement learning framework for autonomous driving using uncertainty-aware expert advice with adaptive thresholds. Epistemic and aleatoric uncertainty trigger expert intervention; commitment-cooldown strategy prevents long-term dependence. CARLA experiments: +5-7% success rate vs IQN baseline.
Two ML students question whether robotics faces a data scarcity problem. After normalizing public datasets, they suspect the real issue is interoperability: heterogeneous schemas, different sensors, incompatible coordinate frames. They ask robotics teams whether they would actually use data from other teams through a unified API.
Shift, a robotics startup, offers free home cleaning services to collect training data for future domestic robots. Business model relies on real-world data acquisition rather than immediate monetization.
Mercedes launches MB.Drive Assist Pro, an urban assisted driving system capable of handling traffic lights and traffic, to compete with Tesla's Full Self-Driving in Europe.
URIEL proposes a selective logging method for tropical forests combining helicopters, robotics and AI to minimize collateral damage. Digital simulation and economic feasibility analysis demonstrate concept viability, but implementation depends on stakeholder integration (industry, governments, certified companies, indigenous populations).
Wall-OSS-0.5 is a 4B VLA from X Square Robot with open training code. Zero-shot evaluation on 17 real-robot tasks: 4 tasks >80% progress, including Rope Tightening (82%). Post fine-tuning: 60.5% average task progress (+17.5pp vs pi0.5). Mixture-of-Transformers architecture with vision-aligned RVQ tokenizer and distributed DMuon optimizer.
Bayesian framework for validating deployment of learned autonomous landing controllers. Uses Bayesian inference to quantify uncertainty about true policy capability beyond empirical metrics (reward, success rate). Experiments with PPO and SAC show empirical optimization overconfidence, while Bayesian inference better calibrates deployment readiness assessment.
A physics-informed machine learning (PIML) framework for thermal modeling of a lunar rover. An adaptive neural network determines 3D finite-difference meshing based on thermal loads, improving accuracy by 50% vs coarse-mesh physics models and 39% vs pure ANN, while being 3x faster than high-fidelity simulations.
RL framework for sim-to-real policy transfer via probabilistic latent embeddings and dynamic adaptation. Uses meta-RL and CMDPs to infer latent environment representation, with distributional RL formulation dynamically adjusting risk levels based on latent context estimation accuracy.
Open-source 7MB self-driving AI model trained on visual and sensor data. Real-time execution on phones and embedded devices without server infrastructure. Learns navigation, lane following, and drift recovery.
LieEDNN introduces Lie group embedded dynamical neural networks to model continuous symmetries on manifolds. The method addresses incompatibility between neural network addition and non-Euclidean geometry through adjoint Lie group actions on Lie algebras. Tested on SE(3) for telescopic manipulator control.
Reachy Mini, Pollen Robotics' humanoid robot, now runs fully locally without cloud dependency. Integrates open-source models (Llama, Whisper) for vision, speech, and motor control. Deployed on embedded hardware.
Neuro-inspired framework for embodied planning and control. The Inverter uses Inverse Learning (IL) to generate multi-step action sequences. Outperforms offline-RL and diffusion-planner baselines on D4RL (+24.2% average) with 100-1000x less inference compute. Application: single-qubit quantum gate synthesis matching GRAPE fidelity at 1000x faster per-gate compute.
Thinking-learning interaction model for autonomous robots in changing environments. Thinking guides learning (change identification, evidence selection, planning), learning improves thinking (knowledge updates, action strategies). Results: recognition accuracy 0.419→0.845, action length 13.0→4.0, evidence selection rate 0.272→0.965.
Call for papers for the 2nd Workshop on Efficient Reasoning at COLM 2026 (October 9). Deadline: July 12, 2026. Topics: multimodal reasoning under efficiency constraints, dataset curation, algorithmic innovations, fast inference (pruning, compression, KV-cache), benchmarks, on-device deployment, safety, real-time applications (healthcare, robotics).
Fast-dDrive is a block-diffusion VLA (Vision-Language-Action) model for autonomous driving. It combines bidirectional refinement within semantic units with strict causal ordering, handles structured JSON outputs, and achieves 12× throughput speedup with SGLang. On nuScenes, L2 error reduced to 0.32m (22% improvement), SOTA on WOD-E2E.
User seeks advice on training bottleneck in robotics imitation learning. Pipeline: 4 RGB cameras 128×128 → frozen ResNet18 → DiT (~50M params, 8 layers) predicting action chunks. A4500 GPU at 20–30% utilization, CPU saturated, ~10 iter/sec. Profiler shows optimizer_step dominant (62.4%).
Alibaba releases Qwen3.7-Max, a proprietary model designed for long-running autonomous agent tasks. It matches Claude Opus 4.6 on benchmarks and outperforms DeepSeek V4 Pro and Kimi K2.6. The model ran autonomously for 35 hours to optimize code for Alibaba's custom chip.
ScenePilot generates critical scenarios for autonomous driving testing via multi-objective reinforcement learning. The framework combines RSS-derived physical feasibility with an AV-risk predictor to target boundary-band scenarios: physically solvable yet causing failures. Results: +6.2 percentage points collision rate on SafeBench while preserving physical validity.
Joint training of autonomous vehicle and 12 pedestrians using MARL (MAPPO) in simulation. SDC reaches 78% goal completion with 14% collision rate vs 35%/33% for rule-based baseline. Jaywalkers (13% of crossings) account for 62% of collisions. Co-training reduces collisions by 30% vs single-agent RL.
Custom RL algorithm NOML for continuous 6-DoF flight control. Combines TD3 with anchor policy (fixed safe action), hierarchical actor (3 independent MLPs pitch→roll→rest), and mirror learning (left-right symmetry). Solves vanilla TD3 oscillation collapse. Open-sourced under Apache 2.0.
Google DeepMind integrates its Genie 3 world model with Street View data: users drop a pin on a map and explore an AI-generated world based on a real place. Street View becomes a strategic training resource for AI agents and robots.
Rerun is an open-source tool to visualize, query, and stream multimodal robotics data for training. Visualization and data management platform for complex robotics projects.