Understanding driver interactions is critical to designing autonomous vehicles to interoperate safely with human-driven cars. We consider the impact of these interactions on the policies drivers employ when navigating unsigned intersections in a driving simulator. The simulator allows the collection of naturalistic decision-making and behavior data in a controlled environment. Using these data, we model the human driver responses as state-based feedback controllers learned via Gaussian Process regression methods. We compute the feedback gain of the controller using a weighted combination of linear and nonlinear priors. We then analyze how the individual gains are reflected in driver behavior. We also assess differences in these controllers across populations of drivers. Our work in data-driven analyses of how drivers determine their policies can facilitate future work in the design of socially responsive autonomy for vehicles.

Generalizing deep reinforcement learning agents to unseen environments remains a significant challenge. One promising solution is Unsupervised Environment Design (UED), a co-evolutionary framework in which a teacher adaptively generates tasks with high learning potential, while a student learns a robust policy from this evolving curriculum. Existing UED methods typically measure learning potential via regret, the gap between optimal and current performance, approximated solely by value-function loss. Building on these approaches, we introduce the transition-prediction error as an additional term in our regret approximation. To capture how training on one task affects performance on others, we further propose a lightweight metric called Co-Learnability. By combining these two measures, we present Transition-aware Regret Approximation with Co-learnability for Environment Design (TRACED). Empirical evaluations show that TRACED produces curricula that improve zero-shot generalization over strong baselines across multiple benchmarks. Ablation studies confirm that the transition-prediction error drives rapid complexity ramp-up and that Co-Learnability delivers additional gains when paired with the transition-prediction error. These results demonstrate how refined regret approximation and explicit modeling of task relationships can be leveraged for sample-efficient curriculum design in UED. Project Page: https://geonwoo.me/traced/

The growth of the tech startup ecosystem in Latin America (LATAM) is driven by innovative entrepreneurs addressing market needs across various sectors. However, these startups encounter unique challenges and risks that require specific management approaches. This paper explores a case study with the Total Addressable Market (TAM), Serviceable Available Market (SAM), and Serviceable Obtainable Market (SOM) metrics within the context of the online food delivery industry in LATAM, serving as a model for valuing startups using the Discounted Cash Flow (DCF) method. By analyzing key emerging powers such as Argentina, Colombia, Uruguay, Costa Rica, Panama, and Ecuador, the study highlights the potential and profitability of AI-driven startups in the region through the development of a ranking of emerging powers in Latin America for tech startup investment. The paper also examines the political, economic, and competitive risks faced by startups and offers strategic insights on mitigating these risks to maximize investment returns. Furthermore, the research underscores the value of diversifying investment portfolios with startups in emerging markets, emphasizing the opportunities for substantial growth and returns despite inherent risks.

Benchmarks play a crucial role in the development and analysis of reinforcement learning (RL) algorithms. We identify that existing benchmarks used for research into open-ended learning fall into one of two categories. Either they are too slow for meaningful research to be performed without enormous computational resources, like Crafter, NetHack and Minecraft, or they are not complex enough to pose a significant challenge, like Minigrid and Procgen. To remedy this, we first present Craftax-Classic: a ground-up rewrite of Crafter in JAX that runs up to 250x faster than the Python-native original. A run of PPO using 1 billion environment interactions finishes in under an hour using only a single GPU and averages 90% of the optimal reward. To provide a more compelling challenge we present the main Craftax benchmark, a significant extension of the Crafter mechanics with elements inspired from NetHack. Solving Craftax requires deep exploration, long term planning and memory, as well as continual adaptation to novel situations as more of the world is discovered. We show that existing methods including global and episodic exploration, as well as unsupervised environment design fail to make material progress on the benchmark. We believe that Craftax can for the first time allow researchers to experiment in a complex, open-ended environment with limited computational resources.

Motion prediction is critical for autonomous off-road driving, however, it presents significantly more challenges than on-road driving because of the complex interaction between the vehicle and the terrain. Traditional physics-based approaches encounter difficulties in accurately modeling dynamic systems and external disturbance. In contrast, data-driven neural networks require extensive datasets and struggle with explicitly capturing the fundamental physical laws, which can easily lead to poor generalization. By merging the advantages of both methods, neuro-symbolic approaches present a promising direction. These methods embed physical laws into neural models, potentially significantly improving generalization capabilities. However, no prior works were evaluated in real-world settings for off-road driving. To bridge this gap, we present PhysORD, a neural-symbolic approach integrating the conservation law, i.e., the Euler-Lagrange equation, into data-driven neural models for motion prediction in off-road driving. Our experiments showed that PhysORD can accurately predict vehicle motion and tolerate external disturbance by modeling uncertainties. It outperforms existing methods both in accuracy and efficiency and demonstrates data-efficient learning and generalization ability in long-term prediction.

It remains a significant challenge to train generally capable agents with reinforcement learning (RL). A promising avenue for improving the robustness of RL agents is through the use of curricula. One such class of methods frames environment design as a game between a student and a teacher, using regret-based objectives to produce environment instantiations (or levels) at the frontier of the student agent's capabilities. These methods benefit from their generality, with theoretical guarantees at equilibrium, yet they often struggle to find effective levels in challenging design spaces. By contrast, evolutionary approaches seek to incrementally alter environment complexity, resulting in potentially open-ended learning, but often rely on domain-specific heuristics and vast amounts of computational resources. In this paper we propose to harness the power of evolution in a principled, regret-based curriculum. Our approach, which we call Adversarially Compounding Complexity by Editing Levels (ACCEL), seeks to constantly produce levels at the frontier of an agent's capabilities, resulting in curricula that start simple but become increasingly complex. ACCEL maintains the theoretical benefits of prior regret-based methods, while providing significant empirical gains in a diverse set of environments. An interactive version of the paper is available at accelagent.github.io.

Accel.AI was founded in September of 2016, our mission is to drive artificial intelligence for social impact initiatives. We focus on integrating AI and social impact through research, consulting, and workshops, on ethical AI development and applied AI engineering. Our target audience includes underrepresented groups, tech companies, including startups and large corporations, governments, and educating individuals experiencing job loss due to automation. We work with companies, professionals, and students around the world.

Principles of AI are a top-down approach to ethics for artificial intelligence (AI). Recently, we have been seeing lists of principles for AI ethics popping up everywhere. They are very useful, not only for AI and its impact but also on a larger social level. Because of AI, people are thinking about ethics in a whole new way: How do we define and digest ethics in order to codify it? Previously I have written an analysis of top-down and bottom-up approaches to ethics for AI, and then we explored the bottom-up method of reinforcement learning for teaching AI ethics.

Spyne a deep tech start-up helping businesses and marketplaces create high-quality product images and videos at scale with AI has raised $7 million in their latest funding round. Led by Accel, the funding round also saw the participation from other marquee investors including Storm Ventures, Smile Group, Pentathlon Ventures, Core91, and prominent founders/CXOs from leading Internet companies. The fresh capital will be invested in acquiring the right talent, bolstering global expansion, including in the US market, and setting up a state-of-the-art computer vision lab for deeper R&D in the space. The brand also intends to expand its technological horizons into the field of AR / VR to build products for metaverse and omniverse. Founded in 2018 by Sanjay Kumar and Deepti Prasad, Spyne develops 100 per cent automatic, industry-first AI image processing products to help large e-commerce marketplaces in the automotive, fashion, and retail industry enhance the visual value of the images without a physical studio.

Machine Learning can be divided into three different types of learning: Unsupervised Learning, Supervised Learning, and Semi-supervised Learning. Unsupervised learning uses information data that is not labeled, that way the machine should work with no guidance according to patterns, similarities, and differences. On the other hand, supervised learning has a presence of a "teacher", who is in charge of training the machine by labeling the data to work with. Next, the machine receives some examples that allow it to produce a correct outcome. But there's a hybrid approach for these types of learning, this Semi-supervised learning works with both labeled and unlabeled data. This method uses a tiny data set of labeled data to train and label the rest of the data with corresponding predictions, finally giving a solution to the problem.