Workshop

Redistribution-based Cost Inference Improves Sparse Safe Offline RL

Safe offline RL typically assumes access to dense per-step cost annotations, but in practice supervisors provide only trajectory-level …

Ebenezer Gelo, Geraud Nangue Tasse, Steven James, Benjamin Rosman

The Goal-Directed Frame for General Agents

Reinforcement learning is often framed around episodic, discounted, or average scalar rewards. While useful, these views miss a core …

Geraud Nangue Tasse, Steven James, Benjamin Rosman

CORDA: A Benchmark for Hierarchical Harm-Centric Moral Reasoning in Large Language Models

The key question in moral judgement is not simply whether someone chooses the “right” answer, but how they decide what …

Siddarth Singh, Victoria Williams, Simon Rosen, Ebenezer Gelo, Helen Sarah Robertson, Ibrahim Suder, Benjamin Rosman, Geraud Nangue Tasse, Steven James

Optimal Task Generalisation in Cooperative Multi-Agent Reinforcement Learning

While task generalisation is widely studied in the context of single-agent reinforcement learning (RL), little research exists in the …

Simon Rosen, Abdel Mfougouon Njupoun, Geraud Nangue Tasse, Steven James, Benjamin Rosman

ROSARL: Reward-Only Safe Reinforcement Learning

An important problem in reinforcement learning is designing agents that learn to solve tasks safely in an environment. A common …

Geraud Nangue Tasse, Tamlin Love, Mark Nemecek, Steven James, Benjamin Rosman

MinePlanner: A Benchmark for Long-Horizon Planning in Large Minecraft Worlds

We propose a new benchmark for planning tasks based on the Minecraft game. Our benchmark contains 45 tasks overall, but also provides …

William Hill, Ireton Liu, Anita De Mello Koch, Damion Harvey, Nishanth Kumar, George Konidaris, Steven James

Counting Reward Automata: Sample Efficient Reinforcement Learning Through the Exploitation of Reward Function Structure

We present counting reward automata—a finite state machine variant capable of modelling any reward function expressible as a …

Tristan Bester, Benjamin Rosman, Steven James, Geraud Nangue Tasse

End-to-End Learning to Follow Language Instructions with Compositional Policies

We develop an end-to-end model for learning to follow language instructions with compositional policies. Our model combines large …

Vanya Cohen, Geraud Nangue Tasse, Nakul Gopalan, Steven James, Raymond Mooney, Benjamin Rosman

Skill Machines: Temporal Logic Composition in Reinforcement Learning

A major challenge in reinforcement learning is specifying tasks in a manner that is both interpretable and verifiable. One common …

Geraud Nangue Tasse, Devon Jarvis, Steven James, Benjamin Rosman

Augmentative Topology Agents For Open-ended Learning

In this work, we tackle the problem of open-ended learning by introducing a method that simultaneously evolves agents and increasingly …

Muhammad Nasir, Steven James, Christopher Cleghorn