Perceptive deep reinforcement learning (DRL) has lead to many recent breakthroughs for complex AI systems leveraging image-based input data. However, training these perceptive DRL-enabled systems remains incredibly memory intensive. In this paper, we begin to address this issue through differentially encoded observation spaces. By reinterpreting stored image-based observations as a video, we leverage lossless differential video encoding schemes to compress the replay buffer without impacting training performance. We evaluate our approach with three state-of-the-art DRL algorithms and find that differential image encoding reduces the memory footprint by as much as 14.2x and 16.7x across tasks from the Atari 2600 benchmark and the DeepMind Control Suite (DMC) respectively. These savings also enable large-scale perceptive DRL that previously required paging between flash and RAM to be run entirely in RAM, improving the latency of DMC tasks by as much as 32%..
Embedded machine learning (ML) on low-power devices, also known as "TinyML," enables intelligent applications on accessible hardware and fosters collaboration across disciplines to solve real-world problems. Its interdisciplinary and practical nature makes embedded ML education appealing, but barriers remain that limit its accessibility, especially in developing countries. Challenges include limited open-source software, courseware, models, and datasets that can be used with globally accessible heterogeneous hardware. Our vision is that with concerted effort and partnerships between industry and academia, we can overcome such challenges and enable embedded ML education to empower developers and researchers worldwide to build locally relevant AI solutions on low-cost hardware, increasing diversity and sustainability in the field. Towards this aim, we document efforts made by the TinyML4D community to scale embedded ML education globally through open-source curricula and introductory workshops co-created by international educators. We conclude with calls to action to further develop modular and inclusive resources and transform embedded ML into a truly global gateway to embedded AI skills development.
Artificial intelligence systems connected to sensor-laden devices are becoming pervasive, which has significant implications for a range of AI risks, including to privacy, the environment, autonomy, and more. In this paper, we provide a comprehensive analysis of the evolution of sensors, the risks they pose by virtue of their material existence in the world, and the impacts of ubiquitous sensing and on-device AI. We propose incorporating sensors into risk management frameworks and call for more responsible sensor and system design paradigms that address risks of such systems. We show through calculative models that current systems prioritize data collection and cost reduction and produce risks that emerge around privacy, surveillance, waste, and power dynamics. We then analyze these risks, highlighting issues of validity, safety, security, accountability, interpretability, and bias. We conclude by advocating for increased attention to the materiality of algorithmic systems, and of on-device AI sensors in particular, and highlight the need for development of a responsible sensor design paradigm that empowers users and communities and leads to a future of increased fairness, accountability and transparency.
The landscape of artificial intelligence (AI) applications has traditionally been dominated by the use of resource-intensive servers centralised in industrialised nations. However, recent years have witnessed the emergence of small, energy-efficient devices for AI applications, a concept known as tiny machine learning (TinyML). We’re most familiar with consumer-facing applications such as Siri, Alexa, and Google Assistant, but the limited cost and small size of such devices allow them to be deployed in the field. For example, the technology has been used to detect mosquito wingbeats and so help prevent the spread of malaria. It’s also been part of the development of low-power animal collars to support conservation efforts.
In this study we will investigate whether we can reduce the barriers to entry for high school robotics through the use of code generation models derived from large language models (LLMs). As such, we aim to raise the abstraction barrier for the development of artificial intelligence algorithms needed to program and control the Romi Robot used in the FIRST Robotics Competition (FRC). To do so we develop a web interface that helps automate the prompt-engineer step and allows students to easily incorporate OpenAI Codex into their workflows.
The high barriers to entry associated with robotics, in particular its high cost, has rendered it inaccessibility for many. In this poster we present our early efforts to begin to address these challenges through edge machine learning (ML). We show how ultra-low-cost robot and computational hardware paired with open-source software and courseware can be leveraged for hands-on education globally and the beginnings of a globally diverse research community.
This paper introduces a standard datasheet template for ML sensors and discusses its essential components inluding: the system's hardware, ML model and dataset attributes, end-to-end performance metrics, and environmental impact. We provide an example datasheet for our own ML sensor and discuss each section in detail. We highlight how these datasheets can facilitate better understanding and utilization of sensor data in ML applications, and we provide objective measures upon which system performance can be evaluated and compared.
Deep reinforcement learning (DRL) is one of the most powerful tools for synthesizing complex robotic behaviors. But training DRL models is incredibly compute and memory intensive, requiring large training datasets and replay buffers to achieve performant results. This poses a challenge for the next generation of field robots that will need to learn on the edge to adapt to their environment. In this paper, we begin to address this issue through observation space quantization. We evaluate our approach using four simulated robot locomotion tasks and two state-of-the-art DRL algorithms, the on-policy Proximal Policy Optimization (PPO) and off-policy Soft Actor-Critic (SAC) and find that observation space quantization reduces overall memory costs by as much as 4.2x without impacting learning performance.
The rise of TinyML has opened up new opportunities for the development of smart, low-power devices in resource-constrained environments. A network of 40 universities has been established over the past two years with the goal of promoting the use of TinyML in developing regions. The members of this network have taught courses at their home institutions and have completed their first research projects covering topics ranging from the diagnosis of respiratory diseases in Rwanda to assistive technology development in Brazil, bee population monitoring in Kenya and estimating the lifespan of the date palm fruit in Saudi Arabia. We suggest three policy recommendations to increase the future impact: first, training and research activities in STI should focus on regional networks; second, the ethics of artificial intelligence must be covered in all activities; and third, we need to support local champions better.
In this viewpoint we propose the ML sensor: a logical framework for developing ML-enabled embedded systems which empowers end users through its privacy-by-design approach. By limiting the data interface, the ML sensor paradigm helps ensure that no user information can be extracted beyond the scope of the sensor’s functionality. Our proposed definition is as follows: An ML sensor is a self-contained, embedded system that utilizes machine learning to process sensor data on-device – logically decoupling data computation from the main application processor and limiting the data access of the wider system to high-level ML model outputs.
The sustained growth of carbon emissions and global waste elicits significant sustainability concerns for our environment's future. The growing Internet of Things (IoT) has the potential to exacerbate this issue. However, an emerging area known as Tiny Machine Learning (TinyML) has the opportunity to help address these environmental challenges through sustainable computing practices. TinyML, the deployment of machine learning (ML) algorithms onto low-cost, low-power microcontroller systems, enables on-device sensor analytics that unlocks numerous always-on ML applications. This article discusses the potential of these TinyML applications to address critical sustainability challenges. Moreover, the footprint of this emerging technology is assessed through a complete life cycle analysis of TinyML systems. From this analysis, TinyML presents opportunities to offset its carbon emissions by enabling applications that reduce the emissions of other sectors. Nevertheless, when globally scaled, the carbon footprint of TinyML systems is not negligible, necessitating that designers factor in environmental impact when formulating new devices. Finally, research directions for enabling further opportunities for TinyML to contribute to a sustainable future are outlined.
As a step toward robust learning pipelines for these constrained robot platforms, we demonstrate how existing state-of-the-art imitation learning pipelines can be modified and augmented to support low-cost, limited hardware. By reducing our model’s observational space, leveraging TinyML to quantize our model, and adjusting the model outputs through post-processing, we are able to learn and deploy successful walking gaits on an 8-DoF, $299 (USD) toy quadruped robot that has reduced actuation and sensor feedback, as well as limited computing resources.
Tiny robot learning lies at the intersection of embedded systems, robotics, and ML, compounding the challenges of these domains. This paper gives a brief survey of the tiny robot learning space, elaborates on key challenges, and proposes promising opportunities for future work in ML system design.
We believe that TinyML has a significant role to play in achieving the SDGs and facilitating scientific research in areas such as environmental monitoring, physics of complex systems and energy management. To broaden access and participation and increase the impact of this new technology, we present an initiative that is creating and supporting a global network of academic institutions working on TinyML in developing countries. We suggest the development of additional open educational resources, South–South academic collaboration and pilot projects of at-scale TinyML solutions aimed at addressing the SDGs.
[TinyMLedu](https://tinymledu.org) is working to build an international coalition of researchers and practitioners advancing TinyML in the developing world, and to develop and share high-quality, open-access educational materials globally.
In this paper, we describe our pedagogical approach to increasing access to applied ML through a four part massive open online course (MOOC) on Tiny Machine Learning (TinyML) produced in collaboration between academia (Harvard University) and industry (Google). We suggest that TinyML, ML on resource-constrained embedded devices, is an attractive means to widen access because TinyML both leverages low-cost and globally accessible hardware, and encourages the development of complete, self-contained applications, from data collection to deployment. We also released the course materials publicly, hoping they will inspire the next generation of ML practitioners and educators and further broaden access to cutting-edge ML technologies.