Boosting Sample Efficiency and Generalization in Multi-agent Reinforcement Learning via Equivariance

Multi-Agent Reinforcement Learning (MARL) struggles with sample inefficiency and poor generalization [1]. These challenges are partially due to a lack of structure or inductive bias in the neural networks typically used in learning the policy. One such form of structure that is commonly observed in multi-agent scenarios is symmetry. The field of Geometric Deep Learning has developed Equivariant Graph Neural Networks (EGNN) that are equivariant (or symmetric) to rotations, translations, and reflections of nodes. Incorporating equivariance has been shown to improve learning efficiency and decrease error [ 2 ]. In this paper, we demonstrate that EGNNs improve the sample efficiency and generalization in MARL. However, we also show that a naive application of EGNNs to MARL results in poor early exploration due to a bias in the EGNN structure. To mitigate this bias, we present Exploration-enhanced Equivariant Graph Neural Networks or E2GN2. We compare E2GN2 to other common function approximators using common MARL benchmarks MPE and SMACv2. E2GN2 demonstrates a significant improvement in sample efficiency, greater final reward convergence, and a 2x-5x gain in over standard GNNs in our generalization tests. These results pave the way for more reliable and effective solutions in complex multi-agent systems.

IMRL: Integrating Visual, Physical, Temporal, and Geometric Representations for Enhanced Food Acquisition

Robotic assistive feeding holds significant promise for improving the quality of life for individuals with eating disabilities. However, acquiring diverse food items under varying conditions and generalizing to unseen food presents unique challenges. Existing methods that rely on surface-level geometric information (e.g., bounding box and pose) derived from visual cues (e.g., color, shape, and texture) often lacks adaptability and robustness, especially when foods share similar physical properties but differ in visual appearance. We employ imitation learning (IL) to learn a policy for food acquisition. Existing methods employ IL or Reinforcement Learning (RL) to learn a policy based on off-the-shelf image encoders such as ResNet-50. However, such representations are not robust and struggle to generalize across diverse acquisition scenarios. To address these limitations, we propose a novel approach, IMRL (Integrated Multi-Dimensional Representation Learning), which integrates visual, physical, temporal, and geometric representations to enhance the robustness and generalizability of IL for food acquisition. Our approach captures food types and physical properties (e.g., solid, semi-solid, granular, liquid, and mixture), models temporal dynamics of acquisition actions, and introduces geometric information to determine optimal scooping points and assess bowl fullness. IMRL enables IL to adaptively adjust scooping strategies based on context, improving the robot's capability to handle diverse food acquisition scenarios. Experiments on a real robot demonstrate our approach's robustness and adaptability across various foods and bowl configurations, including zero-shot generalization to unseen settings. Our approach achieves improvement up to $35\%$ in success rate compared with the best-performing baseline.

NAVINACT: Combining Navigation and Imitation Learning for Bootstrapping Reinforcement Learning

Reinforcement Learning (RL) has shown remarkable progress in simulation environments, yet its application to real-world robotic tasks remains limited due to challenges in exploration and generalisation. To address these issues, we introduce NAVINACT, a framework that chooses when the robot should use classical motion planning-based navigation and when it should learn a policy. To further improve the efficiency in exploration, we use imitation data to bootstrap the exploration. NAVINACT dynamically switches between two modes of operation: navigating to a waypoint using classical techniques when away from the objects and reinforcement learning for fine-grained manipulation control when about to interact with objects. NAVINACT consists of a multi-head architecture composed of ModeNet for mode classification, NavNet for waypoint prediction, and InteractNet for precise manipulation. By combining the strengths of RL and Imitation Learning (IL), NAVINACT improves sample efficiency and mitigates distribution shift, ensuring robust task execution. We evaluate our approach across multiple challenging simulation environments and real-world tasks, demonstrating superior performance in terms of adaptability, efficiency, and generalization compared to existing methods. In both simulated and real-world settings, NAVINACT demonstrates robust performance. In simulations, NAVINACT surpasses baseline methods by 10-15\% in training success rates at 30k samples and by 30-40\% during evaluation phases. In real-world scenarios, it demonstrates a 30-40\% higher success rate on simpler tasks compared to baselines and uniquely succeeds in complex, two-stage manipulation tasks. Datasets and supplementary materials can be found on our website: {https://raaslab.org/projects/NAVINACT/}.

Is Generative Communication between Embodied Agents Good for Zero-Shot ObjectNav?

In Zero-Shot ObjectNav, an embodied ground agent is expected to navigate to a target object specified by a natural language label without any environment-specific fine-tuning. This is challenging, given the limited view of a ground agent and its independent exploratory behavior. To address these issues, we consider an assistive overhead agent with a bounded global view alongside the ground agent and present two coordinated navigation schemes for judicious exploration. We establish the influence of the Generative Communication (GC) between the embodied agents equipped with Vision-Language Models (VLMs) in improving zero-shot ObjectNav, achieving a 10% improvement in the ground agent's ability to find the target object in comparison with an unassisted setup in simulation. We further analyze the GC for unique traits quantifying the presence of hallucination and cooperation. In particular, we identify a unique trait of "preemptive hallucination" specific to our embodied setting, where the overhead agent assumes that the ground agent has executed an action in the dialogue when it is yet to move. Finally, we conduct real-world inferences with GC and showcase qualitative examples where countering pre-emptive hallucination via prompt finetuning improves real-world ObjectNav performance.

Fast k-connectivity Restoration in Multi-Robot Systems for Robust Communication Maintenance

Maintaining a robust communication network plays an important role in the success of a multi-robot team jointly performing an optimization task. A key characteristic of a robust cooperative multi-robot system is the ability to repair the communication topology in the case of robot failure. In this paper, we focus on the Fast k-connectivity Restoration (FCR) problem, which aims to repair a network to make it k-connected with minimum robot movement. We develop a Quadratically Constrained Program (QCP) formulation of the FCR problem, which provides a way to optimally solve the problem, but cannot handle large instances due to high computational overhead. We therefore present a scalable algorithm, called EA-SCR, for the FCR problem using graph theoretic concepts. By conducting empirical studies, we demonstrate that the EA-SCR algorithm performs within 10 percent of the optimal while being orders of magnitude faster. We also show that EA-SCR outperforms existing solutions by 30 percent in terms of the FCR distance metric.

LAVA: Long-horizon Visual Action based Food Acquisition

Robotic Assisted Feeding (RAF) addresses the fundamental need for individuals with mobility impairments to regain autonomy in feeding themselves. The goal of RAF is to use a robot arm to acquire and transfer food to individuals from the table. Existing RAF methods primarily focus on solid foods, leaving a gap in manipulation strategies for semi-solid and deformable foods. This study introduces Long-horizon Visual Action (LAVA) based food acquisition of liquid, semisolid, and deformable foods. Long-horizon refers to the goal of "clearing the bowl" by sequentially acquiring the food from the bowl. LAVA employs a hierarchical policy for long-horizon food acquisition tasks. The framework uses high-level policy to determine primitives by leveraging ScoopNet. At the mid-level, LAVA finds parameters for primitives using vision. To carry out sequential plans in the real world, LAVA delegates action execution which is driven by Low-level policy that uses parameters received from mid-level policy and behavior cloning ensuring precise trajectory execution. We validate our approach on complex real-world acquisition trials involving granular, liquid, semisolid, and deformable food types along with fruit chunks and soup acquisition. Across 46 bowls, LAVA acquires much more efficiently than baselines with a success rate of 89 +/- 4% and generalizes across realistic plate variations such as different positions, varieties, and amount of food in the bowl. Code, datasets, videos, and supplementary materials can be found on our website.

Adaptive Visual Imitation Learning for Robotic Assisted Feeding Across Varied Bowl Configurations and Food Types

In this study, we introduce a novel visual imitation network with a spatial attention module for robotic assisted feeding (RAF). The goal is to acquire (i.e., scoop) food items from a bowl. However, achieving robust and adaptive food manipulation is particularly challenging. To deal with this, we propose a framework that integrates visual perception with imitation learning to enable the robot to handle diverse scenarios during scooping. Our approach, named AVIL (adaptive visual imitation learning), exhibits adaptability and robustness across different bowl configurations in terms of material, size, and position, as well as diverse food types including granular, semi-solid, and liquid, even in the presence of distractors. We validate the effectiveness of our approach by conducting experiments on a real robot. We also compare its performance with a baseline. The results demonstrate improvement over the baseline across all scenarios, with an enhancement of up to 2.5 times in terms of a success metric. Notably, our model, trained solely on data from a transparent glass bowl containing granular cereals, showcases generalization ability when tested zero-shot on other bowl configurations with different types of food.

Towards Efficient Risk-Sensitive Policy Gradient: An Iteration Complexity Analysis

Reinforcement Learning (RL) has shown exceptional performance across various applications, enabling autonomous agents to learn optimal policies through interaction with their environments. However, traditional RL frameworks often face challenges in terms of iteration complexity and robustness. Risk-sensitive RL, which balances expected return and risk, has been explored for its potential to yield probabilistically robust policies, yet its iteration complexity analysis remains underexplored. In this study, we conduct a thorough iteration complexity analysis for the risk-sensitive policy gradient method, focusing on the REINFORCE algorithm and employing the exponential utility function. We obtain an iteration complexity of $\mathcal{O}(\epsilon^{-2})$ to reach an $\epsilon$-approximate first-order stationary point (FOSP). We investigate whether risk-sensitive algorithms can achieve better iteration complexity compared to their risk-neutral counterparts. Our theoretical analysis demonstrates that risk-sensitive REINFORCE can have a reduced number of iterations required for convergence. This leads to improved iteration complexity, as employing the exponential utility does not entail additional computation per iteration. We characterize the conditions under which risk-sensitive algorithms can achieve better iteration complexity. Our simulation results also validate that risk-averse cases can converge and stabilize more quickly after approximately half of the episodes compared to their risk-neutral counterparts.

Beyond Joint Demonstrations: Personalized Expert Guidance for Efficient Multi-Agent Reinforcement Learning

Multi-Agent Reinforcement Learning (MARL) algorithms face the challenge of efficient exploration due to the exponential increase in the size of the joint state-action space. While demonstration-guided learning has proven beneficial in single-agent settings, its direct applicability to MARL is hindered by the practical difficulty of obtaining joint expert demonstrations. In this work, we introduce a novel concept of personalized expert demonstrations, tailored for each individual agent or, more broadly, each individual type of agent within a heterogeneous team. These demonstrations solely pertain to single-agent behaviors and how each agent can achieve personal goals without encompassing any cooperative elements, thus naively imitating them will not achieve cooperation due to potential conflicts. To this end, we propose an approach that selectively utilizes personalized expert demonstrations as guidance and allows agents to learn to cooperate, namely personalized expert-guided MARL (PegMARL). This algorithm utilizes two discriminators: the first provides incentives based on the alignment of policy behavior with demonstrations, and the second regulates incentives based on whether the behavior leads to the desired objective. We evaluate PegMARL using personalized demonstrations in both discrete and continuous environments. The results demonstrate that PegMARL learns near-optimal policies even when provided with suboptimal demonstrations, and outperforms state-of-the-art MARL algorithms in solving coordinated tasks. We also showcase PegMARL's capability to leverage joint demonstrations in the StarCraft scenario and converge effectively even with demonstrations from non-co-trained policies.

REBEL: A Regularization-Based Solution for Reward Overoptimization in Reinforcement Learning from Human Feedback

In this work, we propose REBEL, an algorithm for sample efficient reward regularization based robotic reinforcement learning from human feedback (RRLHF). Reinforcement learning (RL) performance for continuous control robotics tasks is sensitive to the underlying reward function. In practice, the reward function often ends up misaligned with human intent, values, social norms, etc., leading to catastrophic failures in the real world. We leverage human preferences to learn regularized reward functions and eventually align the agents with the true intended behavior. We introduce a novel notion of reward regularization to the existing RRLHF framework, which is termed as agent preferences. So, we not only consider human feedback in terms of preferences, we also propose to take into account the preference of the underlying RL agent while learning the reward function. We show that this helps to improve the over-optimization associated with the design of reward functions in RL. We experimentally show that REBEL exhibits up to 70% improvement in sample efficiency to achieve a similar level of episodic reward returns as compared to the state-of-the-art methods such as PEBBLE and PEBBLE+SURF.