Skip to main content
Search

# Reinforcement Learning

97 related articles

Subscribe to updates
  1. arXiv

    Two policies for one rendezvous: a learned mixture guidance law meets 10 m and 1 m/s on all 600 multi-revolution low-thrust test missions

    Multi-revolution low-thrust rendezvous must combine a long orbital transfer with meter-level terminal control, and a single policy struggles to cover both state scales. This guidance law has a far-range expert read modified equinoctial elements and a terminal expert read the relative state in the target's local frame, while a temporal-attention gate blends their thrust commands continuously. On 600 test missions it completed every rendezvous at 10 m and 1 m/s tolerances, cutting mean flight time by 50.6% and velocity increment by 44.5% relative to 4 km distance-based hard switching.

  2. arXiv

    Nonlinear system identification stops picking excitation signals: a reinforcement learning agent learns to fill the state space itself

    Excitation design for nonlinear system identification has relied on predefined signal classes and repeated optimization, and classical design mainly minimized parameter variance for linear models. Here space-filling input design is recast as a reinforcement learning problem in which an agent issues amplitude-bounded inputs step by step from system observations, with no input class to choose and no frequency band to tune. On two nonlinear benchmarks, a mass-spring-damper with nonlinear stiffness and a Van der Pol oscillator, the learned policy gives the best overall space-filling performance under the same amplitude bound and distributes samples more uniformly than a state-of-the-art space-filling multisine design.

  3. Science Robotics

    A 5-tonne skid-steer loader learns two-wheel acrobatics: reinforcement learning policies swing, balance and cross bumps on the real machine

    A 5-tonne John Deere 332G skid-steer loader trained with reinforcement learning in simulation and transferred to the real machine can rock itself up onto its front wheels, balance on two wheels and track velocity commands. Against a model predictive control baseline, the two perform comparably at static balancing and velocity tracking, but the baseline loses balance beyond a target speed of 3.5 and cannot cross 75-millimetre undulations, while the learned policy can. The pipeline needs no prior machine model; an actuator network trained on measured data closes the sim-to-real gap.

  4. arXiv

    Video models learn to check the evidence before reasoning: EVAThink hits 63.5% average at 32 frames with 6.7K samples and about 62% shorter output than always reasoning

    EVAThink turns whether a sampled set of frames is enough to answer directly into a training signal: for the same question and answer it builds evidence-dense and evidence-sparse views of a video and keeps only pairs where direct answering degrades on the sparse view but explicit reasoning recovers accuracy, then uses two-stage reinforcement learning to teach the model to switch between direct answering and reasoning. Trained on only 6.7K samples, it reaches 63.5% and 65.6% average across six video understanding benchmarks under fixed 32- and 64-frame budgets, both above the Qwen3-VL-8B-Instruct base; at 64 frames the adaptive mode beats direct-only and think-only by 2.3 and 3.6 percentage points and cuts output length by about 62% relative to think-only.

  5. Periodic trains Neon on its own lab data: 55.3% success on FrontierXRD at lower cost than frontier models

    Periodic trained Periodic Neon on X-ray diffraction data from its own labs, reaching a 55.3% success rate on FrontierXRD, its hardest internal evaluation set, against an initial 2.7% for Kimi K2.6, the open-weight model it post-trained. The evaluation covers 134 samples from its labs that human experts take hours to resolve because of their complexity and ambiguity; Neon surpasses frontier models at a lower cost per analysis and is now deployed in its labs analyzing experiments in the search for better superconductors and magnets.

  6. bioRxiv

    REXzyme translates a chemical reaction into an enzyme sequence, and its formate dehydrogenase design C20 reaches 2.5 times wild-type activity

    Enzyme engineering has relied on directed evolution, which screens thousands of variants even though fewer than 5% of single mutations are beneficial. REXzyme recasts enzyme design as translation: an encoder reads a reaction written in SMILES and a decoder generates an amino acid sequence token by token, after training on about 21 million reaction-enzyme pairs. Steered toward the reverse reaction of formate dehydrogenase, the reduction of CO2 to formate, all 20 designs generated after family-specific fine-tuning and reinforcement-learning alignment expressed as soluble proteins and showed activity, and design C20 reached a specific activity of 80.8 mU/mg, 2.5 times that of wild-type CtFDH.

  7. A trillion-parameter Periodic Neon moves into the physical lab: post-trained on in-house diffraction data, it beats Claude Fable 5.1 and GPT-6 Astra on analysis

    Periodic Neon is a trillion-parameter model post-trained with midtraining and reinforcement learning on data from the company's own labs and deployed to its high-throughput physical labs in Menlo Park to analyze X-ray diffraction results. It outperforms Claude Fable 5.1 and GPT-6 Astra on diffraction analysis, and inference is optimized so the model can be deployed at scale. Its training data comes from physical experiments rather than elastically scalable digital environments.

  8. Hugging Face Blog

    A 0.8B on-device assistant: BotPsy beats a model 2.5 times its size on five of six tool-use benchmarks

    Tether AI Research released BotPsy, a 0.8B-parameter model built on Qwen3.5-0.8B for function calling and agentic workflows that runs on-device. Trained through supervised fine-tuning, multi-teacher on-policy distillation and two reinforcement-learning rounds, it lifts BFCL v4 from its base's 23.4 to 44.0 and τ-bench from 23.2 to 51.5, and beats Qwen3.5-2B-Think, a model 2.5 times its size, on five of six tool-use and agentic benchmarks, trailing only on Claw-Eval (46.5 vs 48.9). It still makes mistakes, targets low-to-medium difficulty tasks and is not intended for high-stakes use.

  9. arXiv

    Soft-arm reinforcement learning drops from three years to three hours: a six-dimensional reduced model trains a tracker with 40% lower error than an RNN policy

    Reinforcement-learning control policies for a pneumatically actuated soft arm can be trained entirely on an offline reduced model instead of through repeated interaction with fragile hardware. Soft robots have lacked a reliable, fast simulator, and trial and error on the hardware itself is costly. The arm's dynamics are reduced to a six-dimensional adiabatic spectral submanifold (aSSM), and a polynomial ordinary differential equation with only 6,924 parameters serves as the training environment; a tracking policy trains in three hours and deploys without fine-tuning, with 40% lower tracking error than otherwise identical policies trained on a 4.8-million-parameter RNN or a 1.7-million-parameter transformer.

  10. bioRxiv

    Why place cells fire in many fields across large spaces: a hierarchical predictive map ties field statistics to the scale of behavior

    In large environments at naturalistic scale, rat and bat hippocampal CA1 place cells often carry multiple irregularly spaced firing fields, whereas the classical single-scale successor representation holds only in small enclosures. The factorised hierarchical successor representation (fHSR) folds one-step actions together with subgoal-directed subroutines into a predictive map, then decomposes the maps of several tasks into a non-negative overcomplete basis that gives each model cell its tuning. On a 48 m rat track and a 200 m bat track the model reproduced the gamma-Poisson overdispersion of field counts, and it predicted a spatial periodicity of population activity whose wavelength is independent of track length, a prediction tested in rat data.