We introduce EvilGenie, a benchmark for reward hacking in programming settings. We source problems from LiveCodeBench and create an environment in which agents can easily reward hack, such as by hardcoding test cases or editing the testing files. We measure reward hacking in three ways: held out unit tests, LLM judges, and test file edit detection. We verify these methods against human review and each other. We find the LLM judge to be highly effective at detecting reward hacking in unambiguous cases, and observe only minimal improvement from the use of held out test cases. In addition to testing many models using Inspect's basic_agent scaffold, we also measure reward hacking rates for three popular proprietary coding agents: OpenAI's Codex, Anthropic's Claude Code, and Google's Gemini CLI Using GPT-5, Claude Sonnet 4, and Gemini 2.5 Pro, respectively. We observe explicit reward hacking by both Codex and Claude Code, and misaligned behavior by all three agents. Our codebase can be found at https://github.com/JonathanGabor/EvilGenie.

Multi-robot flocking possesses extraordinary advantages over a single-robot system in diverse domains, but it is challenging to ensure safe and optimal performance in congested environments. Hence, this paper is focused on the investigation of distributed optimal flocking control for multiple robots in crowded environments. A heuristic predictive control solution is proposed based on a Gibbs Random Field (GRF), in which bio-inspired potential functions are used to characterize robot-robot and robot-environment interactions. The optimal solution is obtained by maximizing a posteriori joint distribution of the GRF in a certain future time instant. A gradient-based heuristic solution is developed, which could significantly speed up the computation of the optimal control. Mathematical analysis is also conducted to show the validity of the heuristic solution. Multiple collision risk levels are designed to improve the collision avoidance performance of robots in dynamic environments. The proposed heuristic predictive control is evaluated comprehensively from multiple perspectives based on different metrics in a challenging simulation environment. The competence of the proposed algorithm is validated via the comparison with the non-heuristic predictive control and two existing popular flocking control methods. Real-life experiments are also performed using four quadrotor UAVs to further demonstrate the efficiency of the proposed design.

The Vehicle Routing Problem with Drones (VRPD) seeks to optimize the routing paths for both trucks and drones, where the trucks are responsible for delivering parcels to customer locations, and the drones are dispatched from these trucks for parcel delivery, subsequently being retrieved by the trucks. Given the NP-Hard complexity of VRPD, numerous heuristic approaches have been introduced. However, improving solution quality and reducing computation time remain significant challenges. In this paper, we conduct a comprehensive examination of heuristic methods designed for solving VRPD, distilling and standardizing them into core elements. We then develop a novel reinforcement learning (RL) framework that is seamlessly integrated with the heuristic solution components, establishing a set of universal principles for incorporating the RL framework with heuristic strategies in an aim to improve both the solution quality and computation speed. This integration has been applied to a state-of-the-art heuristic solution for VRPD, showcasing the substantial benefits of incorporating the RL framework. Our evaluation results demonstrated that the heuristic solution incorporated with our RL framework not only elevated the quality of solutions but also achieved rapid computation speeds, especially when dealing with extensive customer locations.