CORA is a teacher-guided reinforcement-learning framework that improves causal jamming detection recall to 0.771. It trains a masked controller offline using a retrospective KPI detector and deploys it causally without labels.
Key points
PPO controller raises mean recall from 0.687 to 0.771 over a fixed default detector on all ten 5G O-RAN testbed configurations.
DQN achieves a comparable F2-score to the best teacher-selected static configuration with 61% fewer false alarms.
Sources
Sequential Recovery Updates for Learned Controllers adds 62 recoveries
Sequential Recovery Updates for Learned Controllers adds 62 new recoveries over 192 withheld perturbations while preserving all base successes. The paper describes a constraint-based entry rule that protects previously successful trajectories.
Key points
Runs on four streams: two motions on each of two humanoid systems.
Removing the entry constraint recovers 33 more conditions but stops eight protected ones from completing.
Sources
A Hybrid Visitor Prediction Framework – MAE reduced 9.82%
A Hybrid Visitor Prediction Framework using Interpretable Temporal features reduces MAE by 9.82% in lake visitation forecasts. The method adds lag features and uses anomaly detection, outperforming IQR-based approaches on multiple regression metrics.
Key points
Hybrid model reduced MAE 9.82%, MSE 3.12%, RMSE 1.76%, MAPE 6.77%.
Anomaly detection hybrid beat IQR hybrid, lowering MAE 9.81%, MSE 34.30%, RMSE 18.99%, MAPE 0.66%.
Sources
Central pivot point control of a body-attached spring-mass runner
Central pivot point (CPP) control restores central-force structure in trunk-ed SLIP models and yields an analytic return map that predicts liftoff velocity with 1.6% mean error and apex position with 0.9% error. The paper also provides a scalar equation that predicts natural gaits within 0.8% and a deadbeat stride controller based on the analytic map. Results are demonstrated on a linear radial template with closed-form stance radius and constant-momentum sweep.
Key points
Analytic return map predicts liftoff velocity with mean error 1.6% and apex position with mean error 0.9%.
Scalar equation predicts natural gaits within 0.8% error.
Sources
Social Humanoid framework deployed on Unitree G1 robot
Social Humanoid is a framework for socially responsive human–humanoid interaction. It is deployed on a Unitree G1 and uses a hierarchical system that combines high-level social reasoning with low-level control to enable a humanoid to interact with a human partner.
Key points
Framework: Social Humanoid for socially responsive human–humanoid interaction
Deployment: Unitree G1 robot with hierarchical system combining high-level social reasoning and low-level control
Sources
AI Music Video Show – alternative watch options announced
AI Music Video Show provides alternative streaming options while X LIVE issues persist. Watch on Roku by searching “AI Music Video Show”. Also available on AppleTV and Roku via AlchemyStream.
Key points
Roku: search "AI Music Video Show"
AppleTV + Roku via AlchemyStream
Sources
- Original post
- musicvideochannel.ai - Linked in the post
Hermes Desktop gateway inside Alan's Workspace launch
Hermes Desktop gateway inside Alan's Workspace launches this weekend, enabling same chats on phone, iPad and any browser. It adds a desktop gateway to Alan's Way and keeps conversations synchronized across devices.
Key points
Hermes Desktop gateway inside Alan's Workspace arriving this weekend
Same conversations available in the Hermes mobile app on phone, iPad and any browser
Sources
Energy-Based Policy Orchestration paper announced at CoRL 2026
Energy-Based Policy Orchestration introduces EPO, an energy-based orchestrator for modular goal-conditioned policies. It learns a composite energy per policy to refine goals and select the lowest-energy policy-goal pair at runtime. Simulations show improved precision-robustness trade-off, and the same frozen policies and energy models are deployed on real humanoid loco-manipulation tasks.
Key points
Authors: Weirui Ye, Yen-Jen Wang, Elvis Hsieh, Zhiyang Dou, Shibo Zhao, Pieter Abbeel, Yilun Du, Leslie Pack Kaelbling, Tomás Lozano-Pérez.
Presented at CoRL 2026 with arXiv preprint arXiv:2611.02546.
Sources
- Original post
- Energy-Based Policy Orchestration - Linked page
