Researchers investigated reinforcement learning (RL) for adaptive traffic signal control at an urban intersection in Kuwait, aiming to mitigate urban traffic congestion. They developed a Proximal Policy Optimization (PPO)-based controller that dynamically adjusts green-phase durations using local traffic states in a realistic simulation environment informed by real-world Kuwaiti traffic data. The controller reduced average vehicle delay by 46% relative to fixed-time control and 34% relative to actuated control, while also lowering per-vehicle CO2 emissions by approximately 23%. Why it matters: This demonstrates a practical, learning-based edge traffic signal control solution for IoT-enabled smart city transportation systems, offering significant improvements in traffic flow and environmental impact for car-dependent cities in the Middle East.
Technology Innovation Institute (TII) has developed AI-powered autonomous drones capable of navigating complex environments at speeds up to 80 km/h using only a camera and IMU sensor. The drones use onboard AI-driven visual odometry and reinforcement learning to adapt to their environment in real time. In direct competition, the TII drone set a best lap time of 4.38s, compared to 6.32s and 5.34s for human pilots. Why it matters: This research demonstrates the potential of AI-powered UAVs to surpass human-operated drones in agility and precision, with applications for the transport of goods and potentially people.
TII's Autonomous Robotics Research Center (ARRC) and NYU's Agile Robotics and Perception Lab have released RLtools, an open-source reinforcement learning library. RLtools achieves a 75x speed-up in training compared to existing libraries, enabling drone controller training on standard computers. It allows training on consumer-grade laptops or directly on microcontrollers, addressing resource efficiency and deployment challenges. Why it matters: This library accelerates the development and deployment of autonomous systems by reducing training time and resource requirements, making advanced AI more accessible.
This study introduces a reinforcement learning (RL) framework using Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC) to optimize the cleaning schedules of photovoltaic panels in arid regions. Applied to a case study in Abu Dhabi, the PPO-based framework demonstrated up to 13% cost savings compared to simulation optimization methods by dynamically adjusting cleaning intervals based on environmental conditions. The research highlights the potential of RL in enhancing the efficiency and reducing the operational costs of solar power generation.
Researchers at MBZUAI introduce "Interactive Video Reasoning," a new paradigm enabling models to actively "think with videos" by performing iterative visual actions to gather and refine evidence. They developed Video CoM, which reasons through a Chain of Manipulations (CoM), and constructed Video CoM Instruct, an 18K instruction tuning dataset for multi-step manipulation reasoning. The model is further optimized via reinforcement learning with reasoning aware Group Relative Policy Optimization (GRPO), achieving strong results across nine video reasoning benchmarks.
Researchers at MBZUAI have introduced Video-R2, a reinforcement learning approach to improve the consistency and visual grounding of reasoning in multimodal language models. Video-R2 combines timestamp-aware supervised fine-tuning with Group Relative Policy Optimization (GRPO) guided by a Temporal Alignment Reward (TAR). The model demonstrates higher Think Answer Consistency (TAC), Video Attention Score (VAS), and accuracy across multiple benchmarks, showing improved temporal alignment and reasoning coherence for video understanding.
A new method is proposed to reduce the verbosity of LLMs in step-by-step reasoning by retaining moderately easy problems during Reinforcement Learning with Verifiable Rewards (RLVR) training. This approach acts as an implicit length regularizer, preventing the model from excessively increasing output length on harder problems. Experiments using Qwen3-4B-Thinking-2507 show the model achieves baseline accuracy with nearly twice shorter solutions.
A new survey paper provides a deep dive into post-training methodologies for Large Language Models (LLMs), analyzing their role in refining LLMs beyond pretraining. It addresses key challenges such as catastrophic forgetting, reward hacking, and inference-time trade-offs, and highlights emerging directions in model alignment, scalable adaptation, and inference-time reasoning. The paper also provides a public repository to continually track developments in this fast-evolving field.