Tim Seyde

Tim Seyde, Peter Werner, Wilko Schwarting et al. · deepmind

While there has been substantial success for solving continuous control with actor-critic methods, simpler critic-only methods such as Q-learning find limited application in the associated high-dimensional action spaces. However, most actor-critic methods come at the cost of added complexity: heuristics for stabilisation, compute requirements and wider hyperparameter search spaces. We show that a simple modification of deep Q-learning largely alleviates these issues. By combining bang-bang action discretization with value decomposition, framing single-agent control as cooperative multi-agent reinforcement learning (MARL), this simple critic-only approach matches performance of state-of-the-art continuous actor-critic methods when learning from features or pixels. We extend classical bandit examples from cooperative MARL to provide intuition for how decoupled critics leverage state information to coordinate joint optimization, and demonstrate surprisingly strong performance across a variety of continuous control tasks.

Ryan Sander, Wilko Schwarting, Tim Seyde et al. · mit

Experience replay plays a crucial role in improving the sample efficiency of deep reinforcement learning agents. Recent advances in experience replay propose using Mixup (Zhang et al., 2018) to further improve sample efficiency via synthetic sample generation. We build upon this technique with Neighborhood Mixup Experience Replay (NMER), a geometrically-grounded replay buffer that interpolates transitions with their closest neighbors in state-action space. NMER preserves a locally linear approximation of the transition manifold by only applying Mixup between transitions with vicinal state-action features. Under NMER, a given transition's set of state action neighbors is dynamic and episode agnostic, in turn encouraging greater policy generalizability via inter-episode interpolation. We combine our approach with recent off-policy deep reinforcement learning algorithms and evaluate on continuous control environments. We observe that NMER improves sample efficiency by an average 94% (TD3) and 29% (SAC) over baseline replay buffers, enabling agents to effectively recombine previous experiences and learn from limited data.

Lianhao Yin, Makram Chahine, Tsun-Hsuan Wang et al.

The cooperation of a human pilot with an autonomous agent during flight control realizes parallel autonomy. We propose an air-guardian system that facilitates cooperation between a pilot with eye tracking and a parallel end-to-end neural control system. Our vision-based air-guardian system combines a causal continuous-depth neural network model with a cooperation layer to enable parallel autonomy between a pilot and a control system based on perceived differences in their attention profiles. The attention profiles for neural networks are obtained by computing the networks' saliency maps (feature importance) through the VisualBackProp algorithm, while the attention profiles for humans are either obtained by eye tracking of human pilots or saliency maps of networks trained to imitate human pilots. When the attention profile of the pilot and guardian agents align, the pilot makes control decisions. Otherwise, the air-guardian makes interventions and takes over the control of the aircraft. We show that our attention-based air-guardian system can balance the trade-off between its level of involvement in the flight and the pilot's expertise and attention. The guardian system is particularly effective in situations where the pilot was distracted due to information overload. We demonstrate the effectiveness of our method for navigating flight scenarios in simulation with a fixed-wing aircraft and on hardware with a quadrotor platform.

Tsun-Hsuan Wang, Wei Xiao, Tim Seyde et al.

The advancement of robots, particularly those functioning in complex human-centric environments, relies on control solutions that are driven by machine learning. Understanding how learning-based controllers make decisions is crucial since robots are often safety-critical systems. This urges a formal and quantitative understanding of the explanatory factors in the interpretability of robot learning. In this paper, we aim to study interpretability of compact neural policies through the lens of disentangled representation. We leverage decision trees to obtain factors of variation [1] for disentanglement in robot learning; these encapsulate skills, behaviors, or strategies toward solving tasks. To assess how well networks uncover the underlying task dynamics, we introduce interpretability metrics that measure disentanglement of learned neural dynamics from a concentration of decisions, mutual information and modularity perspective. We showcase the effectiveness of the connection between interpretability and disentanglement consistently across extensive experimental analysis.

Rohin Manvi, Joey Hong, Tim Seyde et al.

Large language models excel at reasoning but lack key aspects of introspection, including anticipating their own success and the computation required to achieve it. Humans use real-time introspection to decide how much effort to invest, when to make multiple attempts, when to stop, and when to signal success or failure. Without this, LLMs struggle to make intelligent meta-cognition decisions. Test-time scaling methods like Best-of-N drive up cost and latency by using a fixed budget of samples regardless of the marginal benefit of each one at any point in generation, and the absence of confidence signals can mislead people, prevent appropriate escalation to better tools, and undermine trustworthiness. Learned verifiers or reward models can provide confidence estimates, but do not enable adaptive inference and add substantial cost by requiring extra models or forward passes. We present ZIP-RC, an adaptive inference method that equips models with zero-overhead inference-time predictions of reward and cost. At every token, ZIP-RC reuses reserved or unused logits in the same forward pass as next-token prediction to output a joint distribution over final reward and remaining length -- no extra models, architecture change, or inference overhead. This full joint distribution is used to compute a sampling utility which is the linear combination of the expected maximum reward, total compute, and latency of set of samples if generated to completion. During inference, we maximize this utility with meta-actions that determine which prefix of tokens to continue or initiate sampling from. On mixed-difficulty mathematical benchmarks, ZIP-RC improves accuracy by up to 12% over majority voting at equal or lower average cost, and traces smooth Pareto frontiers between quality, compute, and latency. By providing real-time reward-cost introspection, ZIP-RC enables adaptive, efficient reasoning.

Tim Seyde, Peter Werner, Wilko Schwarting et al. · deepmind

Recent reinforcement learning approaches have shown surprisingly strong capabilities of bang-bang policies for solving continuous control benchmarks. The underlying coarse action space discretizations often yield favourable exploration characteristics while final performance does not visibly suffer in the absence of action penalization in line with optimal control theory. In robotics applications, smooth control signals are commonly preferred to reduce system wear and energy efficiency, but action costs can be detrimental to exploration during early training. In this work, we aim to bridge this performance gap by growing discrete action spaces from coarse to fine control resolution, taking advantage of recent results in decoupled Q-learning to scale our approach to high-dimensional action spaces up to dim(A) = 38. Our work indicates that an adaptive control resolution in combination with value decomposition yields simple critic-only algorithms that yield surprisingly strong performance on continuous control tasks.

Alexander Amini, Anna Banaszak, Harold Benoit et al.

We present LFM2, a family of Liquid Foundation Models designed for efficient on-device deployment and strong task capabilities. Using hardware-in-the-loop architecture search under edge latency and memory constraints, we obtain a compact hybrid backbone that combines gated short convolutions with a small number of grouped query attention blocks, delivering up to 2x faster prefill and decode on CPUs compared to similarly sized models. The LFM2 family covers 350M-8.3B parameters, including dense models (350M, 700M, 1.2B, 2.6B) and a mixture-of-experts variant (8.3B total, 1.5B active), all with 32K context length. LFM2's training pipeline includes a tempered, decoupled Top-K knowledge distillation objective that avoids support mismatch; curriculum learning with difficulty-ordered data; and a three-stage post-training recipe of supervised fine-tuning, length-normalized preference optimization, and model merging. Pre-trained on 10-12T tokens, LFM2 models achieve strong results across diverse benchmarks; for example, LFM2-2.6B reaches 79.56% on IFEval and 82.41% on GSM8K. We further build multimodal and retrieval variants: LFM2-VL for vision-language tasks, LFM2-Audio for speech, and LFM2-ColBERT for retrieval. LFM2-VL supports tunable accuracy-latency tradeoffs via token-efficient visual processing, while LFM2-Audio separates audio input and output pathways to enable real-time speech-to-speech interaction competitive with models 3x larger. LFM2-ColBERT provides a low-latency encoder for queries and documents, enabling high-performance retrieval across multiple languages. All models are released with open weights and deployment packages for ExecuTorch, llama.cpp, and vLLM, making LFM2 a practical base for edge applications that need fast, memory-efficient inference and strong task capabilities.

Tim Seyde, Igor Gilitschenski, Wilko Schwarting et al.

Reinforcement learning (RL) for continuous control typically employs distributions whose support covers the entire action space. In this work, we investigate the colloquially known phenomenon that trained agents often prefer actions at the boundaries of that space. We draw theoretical connections to the emergence of bang-bang behavior in optimal control, and provide extensive empirical evaluation across a variety of recent RL algorithms. We replace the normal Gaussian by a Bernoulli distribution that solely considers the extremes along each action dimension - a bang-bang controller. Surprisingly, this achieves state-of-the-art performance on several continuous control benchmarks - in contrast to robotic hardware, where energy and maintenance cost affect controller choices. Since exploration, learning,and the final solution are entangled in RL, we provide additional imitation learning experiments to reduce the impact of exploration on our analysis. Finally, we show that our observations generalize to environments that aim to model real-world challenges and evaluate factors to mitigate the emergence of bang-bang solutions. Our findings emphasize challenges for benchmarking continuous control algorithms, particularly in light of potential real-world applications.

Wilko Schwarting, Tim Seyde, Igor Gilitschenski et al.

Learning competitive behaviors in multi-agent settings such as racing requires long-term reasoning about potential adversarial interactions. This paper presents Deep Latent Competition (DLC), a novel reinforcement learning algorithm that learns competitive visual control policies through self-play in imagination. The DLC agent imagines multi-agent interaction sequences in the compact latent space of a learned world model that combines a joint transition function with opponent viewpoint prediction. Imagined self-play reduces costly sample generation in the real world, while the latent representation enables planning to scale gracefully with observation dimensionality. We demonstrate the effectiveness of our algorithm in learning competitive behaviors on a novel multi-agent racing benchmark that requires planning from image observations. Code and videos available at https://sites.google.com/view/deep-latent-competition.

Tim Seyde, Wilko Schwarting, Sertac Karaman et al.

Learning complex robot behaviors through interaction requires structured exploration. Planning should target interactions with the potential to optimize long-term performance, while only reducing uncertainty where conducive to this objective. This paper presents Latent Optimistic Value Exploration (LOVE), a strategy that enables deep exploration through optimism in the face of uncertain long-term rewards. We combine latent world models with value function estimation to predict infinite-horizon returns and recover associated uncertainty via ensembling. The policy is then trained on an upper confidence bound (UCB) objective to identify and select the interactions most promising to improve long-term performance. We apply LOVE to visual robot control tasks in continuous action spaces and demonstrate on average more than 20% improved sample efficiency in comparison to state-of-the-art and other exploration objectives. In sparse and hard to explore environments we achieve an average improvement of over 30%.

Tim Seyde, Jan Carius, Ruben Grandia et al.

Locomotion planning for legged systems requires reasoning about suitable contact schedules. The contact sequence and timings constitute a hybrid dynamical system and prescribe a subset of achievable motions. State-of-the-art approaches cast motion planning as an optimal control problem. In order to decrease computational complexity, one common strategy separates footstep planning from motion optimization and plans contacts using heuristics. In this paper, we propose to learn contact schedule selection from high-level task descriptors using Bayesian optimization. A bi-level optimization is defined in which a Gaussian process model predicts the performance of trajectories generated by a motion planning nonlinear program. The agent, therefore, retains the ability to reason about suitable contact schedules, while explicit computation of the corresponding gradients is avoided. We delineate the algorithm in its general form and provide results for planning single-legged hopping. Our method is capable of learning contact schedule transitions that align with human intuition. It performs competitively against a heuristic baseline in predicting task appropriate contact schedules.