In this episode, Jeremiah reveals the twists and lessons of his PhD experience - from navigating advisor changes and overcoming research bottlenecks, to leading teams and mentoring students.

VR teleoperation in robotics. One of Jeremiah’s projects involved a MetaQuest VR headset to tele-operate a Franka robot arm, a process that is foundational to imitation learning in robotics. This setup translates the user’s hand movements, tracked by the headset, into precise actions for the robot’s end effector, allowing for the collection of demonstration data. While he describes the initial experience of controlling a robot with his own hands as feeling very powerful, the novelty wears off when faced with the repetitive reality of data collection. Performing a simple task hundreds of times can lead to fatigue and even motion sickness-like symptoms such as headaches and disorientation. To combat this, researchers have found a clever workaround: wearing the headset around the neck. The sensors can still track hand movements without requiring the user to be fully immersed in VR. However, significant user experience bottlenecks remain. One major issue is lag in the augmented reality passthrough; if the visual feed doesn’t keep up with head movements, it can be extremely disorienting. Another challenge is display resolution, which is limited by the physical size of LEDs. Current technology cannot produce pixels small enough to create a truly crisp image so close to the eyes. Furthermore, the lack of haptic feedback is a major limitation. Without the sense of touch, performing dexterous manipulation tasks that humans do effortlessly, like opening a soda can, becomes incredibly difficult for a robot.

The data bottleneck and distribution shift. Modern robotics is striving to emulate the success of large language models (LLMs) like GPT-4, but it faces a critical obstacle: a massive data bottleneck. LLMs are trained on enormous datasets scraped from the internet, a resource that simply doesn’t exist for robotics. Few people own robots, and the data from those that do are rarely shared publicly. Consequently, most robot data must be collected manually through human tele-operation. While tech giants like Google and Tesla employ contractors to brute force data collection around the clock, this approach is expensive and not scalable. To overcome this, researchers are exploring innovative methods like learning from human video data, leveraging simulation, and, most central to Jeremiah’s work, augmenting existing robot data to make it more diverse and useful. This ties into another fundamental challenge in machine learning: distribution shift. This occurs when the environment a robot is trained in differs from the one it’s deployed in. Jeremiah provides a striking example from a conference: a robot trained for months on tables of a specific height failed completely when moved to a table of a different height. His own research tackles a similar problem with camera viewpoints, where a policy can fail if the camera is moved even slightly. This fragility raises questions about the true intelligence of these systems. Jeremiah argues that the most direct way to build more robust robots is to train them on more diverse data, which is precisely what his work on data augmentation aims to achieve.

A disseration on data augmentation. Jeremiah’s PhD dissertation is focused on robot data augmentation, a field he believes is key to unlocking the next wave of progress in robotics. His vision extends beyond academia; he hopes to apply his research in industry to help companies developing general-purpose robots scale their data collection efforts efficiently. He introduces the concept of performance per dollar, aiming to develop methods that maximize a robot’s learning from each piece of collected data. Since gathering robot data requires significant resources—unlike scraping data from the web—it’s crucial to make every demonstration as valuable as possible. His work involves leveraging computational techniques to augment manually collected data, increasing its diversity and usefulness without requiring additional human effort. However, he acknowledges that quantifying the impact of this work is a challenge in itself. The robotics field currently lacks standardized and reproducible evaluation methods, making it difficult to compare the performance of different models from one lab to another. The processes are often opaque, with little transparency about what constitutes a success or failure, or which tasks were conveniently left out of a report. This lack of a standardized benchmark makes it hard to definitively measure improvements. Despite these challenges, Jeremiah is committed to developing techniques that will help bridge the gap between limited data and the creation of robust, general-purpose robots that can adapt to the complexities of the real world.

Experiences at NASA JPL. Jeremiah’s journey into robotics research began with a seven-month co-op at NASA’s Jet Propulsion Laboratory (JPL) during his undergraduate studies. He worked on the Mars Perseverance Rover, an experience he describes as nothing short of amazing. His specific project involved the percussive mechanism of the rover’s coring drill, which is designed to collect rock samples from the Martian surface. He was part of the team responsible for the hammer that strikes the back of the drill, a critical component for penetrating hard rock. He vividly recalls his first day, which included a tour of the clean room where the rover was being assembled and the Mars yard, a simulated Martian landscape used for testing a full-scale replica of the Curiosity rover. While his daily tasks revolved around mechanical design in CAD software, the true learning came from grappling with the extreme requirements of space engineering. He cites the example of thermal gradients: in the vacuum of space, the side of a satellite facing the sun can be hundreds of degrees hotter than the shaded side, causing immense thermal expansion that can crack the structure. This was a problem he had never considered before. However, the most pivotal moment came from observing the controls engineer on his team. This engineer wrote the algorithms that controlled the drill, including an adaptive coring program that optimized speed and energy use. Witnessing the power of intelligent control systems inspired Jeremiah to shift his focus and pursue a graduate degree in controls and robotics.

Lessons from the racetrack. Long before he was navigating the complexities of robot learning, Jeremiah was managing a 50-person team as the co-founder and captain of a Formula SAE team in his freshman and sophomore years of college. This student engineering competition challenges teams to build a single-seater race car from scratch — an undertaking that consumed nearly all his time before his internship at NASA. This experience, he says, taught him one of the most crucial lessons in engineering: success is often more about project and time management than about creating a super fancy design. He emphasizes that implementation is everything. It’s far better to have a simple, working system than a complex, unfinished one. This philosophy has become a cornerstone of his approach to his PhD research. In the fast-paced world of academic publishing, a brilliant idea is useless if it cannot be implemented, tested, and validated in a timely manner. He notes that it’s easy for researchers to get stuck in their own heads, perpetually brainstorming without ever producing tangible results. His time leading the race car team instilled in him a practical, hands-on mindset focused on execution and delivering a finished product, a skill that has proven invaluable in his academic career. The pressure of building a functional race car from the ground up taught him to prioritize what is essential and to consistently move forward, one implemented component at a time.

Qualities of great researchers. When asked about the qualities that define a great researcher, Jeremiah points to two key attributes: a relentless work ethic and a deep commitment to collaboration. He believes that consistent effort is the bedrock of any successful research endeavor. However, he places even greater emphasis on the power of working with others. Collaboration, in his view, is not just about generating more ideas or solving problems faster; it’s about creating a network of stakeholders who are all invested in a project’s success. This shared investment creates a powerful form of positive pressure and accountability. When you know that others are depending on the project’s outcome, it motivates you to execute and deliver. Furthermore, collaborators bring their unique expertise, resources, and perspectives, creating a support system that can help overcome obstacles that would be insurmountable for a solo researcher. Jeremiah’s advocacy for collaboration stems from his own experiences during the pandemic, a period of isolation that he found to be highly unproductive. He contrasts the collaborative mindset with that of a bad researcher, who he describes as someone who is perpetually lost in their own thoughts and never actually implements anything. For Jeremiah, research is not a solitary pursuit but a team sport, where shared goals and collective effort lead to the most significant breakthroughs.

Building community through collaboration. Jeremiah’s belief in collaboration is more than just a philosophy; it’s a practice he actively cultivates. His desire to build a strong research community was forged during the isolation of the COVID-19 pandemic, a time he found particularly unproductive. Realizing that he thrives in a collaborative environment, he has taken concrete steps to foster connections among his peers. One of his most successful initiatives is the creation of a VLA (Vision-Language-Action) and Embodied AI reading group. This informal, open-invite group meets weekly to discuss a recent paper, providing a much-needed forum for interaction and intellectual exchange. The reading group serves several important functions. First, it gets PhD students out from behind their desks and encourages them to talk to one another, counteracting the natural tendency to work in isolation. Second, it ensures that everyone stays current with the latest research in a rapidly evolving field. With so many papers being published, it can be easy to fall behind, especially when focused on a specific project deadline. The group provides structure and accountability for continuous learning. Finally, and perhaps most importantly, it helps build a sense of community and a network of supportive colleagues. By creating spaces for students to meet, share ideas, and learn from each other, Jeremiah is actively building the collaborative ecosystem that he believes is essential for both personal productivity and cutting-edge research.

The balance between reading and experimenting. In research, there is a natural tension between studying the work of others and creating your own. Jeremiah stresses the importance of finding a healthy balance between reading papers and hands-on experimentation. He warns that focusing exclusively on reading can be a trap for new PhD students. Immersing yourself in the vast body of existing literature can lead to a sense of paralysis, where it feels like every good idea has already been explored and there are no new research gaps to fill. This can stifle creativity and prevent a researcher from ever starting their own experiments. To avoid this pitfall, Jeremiah advocates for a cyclical process: read, think, and then program or run experiments. This cycle ensures that learning from the state of the art directly informs practical implementation, and the results of those experiments, in turn, guide the direction of future reading and thinking. He acknowledges that the ideal balance between these activities is not fixed; it is highly personal and depends on the specific stage of a project. When a deadline is approaching, the focus will naturally shift almost entirely to running experiments and analyzing results. Conversely, at the beginning of a new project or during a brainstorming phase, a researcher might spend a significant amount of time reading to understand the landscape and identify promising avenues for exploration. The key is to never let one activity dominate for too long, ensuring a dynamic interplay between theory and practice.

Prioritizing mental health in academia. The path of a PhD student is notoriously demanding, and Jeremiah is candid about the importance of prioritizing mental health. He recalls the isolation of the pandemic as a particularly challenging time, and he credits his hobbies with helping him navigate the stress and uncertainty. For him, rock climbing became a vital outlet, providing not only a physical challenge but also a strong sense of community. He started climbing indoors and soon expanded to outdoor climbing trips across the Southeast, finding that getting outside and being active was a powerful antidote to the pressures of graduate school. In addition to climbing, he maintains a range of other fitness-related hobbies, including running, yoga, and hiking. Jeremiah shares a crucial insight he has gained: it is essential to have something in your life that is going well when your research is not. If work is the only thing in your life, and that work hits a roadblock — which is inevitable in research — it can feel overwhelming and catastrophic. Hobbies, especially physical ones, provide a reliable source of positive feedback and accomplishment. As he puts it, he never regrets going for a workout and always feels better afterward. This separation of identity from work allows for greater resilience in the face of academic ups and downs. By investing in his well-being outside of the lab, he ensures that he has the mental and emotional fortitude to tackle the long and often frustrating journey of a PhD.

Navigating an advisor change. Changing advisors during a PhD is one of the most difficult and stressful experiences a student can face, and it’s a path Jeremiah has walked himself. He speaks with empathy about the process, noting that other students often seek his advice when they find themselves in this painful situation. It is never a pretty thing, he says, and is always fraught with stress and uncertainty. Based on his experience, he offers three key pieces of advice for students navigating this challenging transition. First, he stresses the importance of staying on the research grind. Even while searching for a new advisor and a new project, it is crucial to keep making progress on your current work. This not only demonstrates commitment and work ethic but also keeps the momentum going. Second, he urges students not to lose hope. Finding a new advisor is possible, and things will eventually be fine. It may extend the timeline of the PhD by a year, but if a doctorate is truly the goal, he believes the extra time will be worth it. Finally, he offers a more controversial piece of advice: consider applying for industry jobs. He has seen peers who decided to leave their PhD program with a master’s degree and were ultimately happy with their decision. Exploring all options can provide a sense of agency and a valuable perspective on what path will lead to the most personal and professional satisfaction.

Choosing between academia and industry. Jeremiah’s decision to pursue a PhD was driven by a deep-seated intellectual curiosity and a desire to engage with more complex mathematical concepts than he felt were utilized in entry-level mechanical engineering roles. While he knew he wanted to attend graduate school, the choice between a master’s program and a PhD was less clear. After much debate with his roommate, who was facing the same decision, Jeremiah landed on a pragmatic strategy: he reasoned it was easier to apply for a PhD program and leave with a master’s degree if he didn’t like it than it was to complete a master’s and then try to get into a PhD program later. Looking back, he admits this might not have been the most robust decision-making process and that he may have benefited from starting with a master’s degree. Nevertheless, the path he chose has ultimately worked out for him. His journey highlights a common crossroads for many aspiring researchers. He had several exciting industry internships, including at NASA and SpaceX, and could have easily pursued a lucrative career in industry. However, his strong desire for a deeper education and his fascination with the complex challenges of robotics and control systems ultimately tipped the scales in favor of academia. His choice underscores the idea that a PhD is not just a career move but a commitment to pushing the boundaries of knowledge, driven by an insatiable curiosity that industry roles might not always satisfy.

A mechanical engineer’s edge in AI. While the current frontier of robotics is dominated by challenges in artificial intelligence, Jeremiah’s background in mechanical engineering gives him a distinct advantage. He is quick to acknowledge that the primary bottleneck in the field today is intelligence, not hardware design. As he puts it, “we can design arbitrarily complex robots, but the barrier is getting them to do intelligent things”. This is why so much research is focused on the intersection of robotics and AI. However, his hands-on engineering skills have proven to be an invaluable asset in his research. His mechanical expertise allows him to implement things faster and more effectively. When a project requires a new physical setup, such as a custom camera mount or a novel task environment, he can design and build it with ease. Furthermore, his deep understanding of how physical systems work makes the low-level aspects of robotics—like debugging hardware and control systems—far less opaque to him. This practical skill set was evident early in his academic career. As an undergrad, he built a 3D printer from scratch, a project he found more engaging than the subsequent material science research he conducted with it. That experience helped clarify his passion for the mechatronics and control systems at the heart of robotics, setting him on his current path.

Advice for aspiring undergraduate researchers. For undergraduates inspired by the incredible feats of modern robots and eager to enter the field, Jeremiah offers clear and practical advice. First and foremost, he emphasizes the importance of getting direct, hands-on experience. He urges students at research universities to actively reach out to labs and graduate students to get involved in a project. In his view, nothing can replace the experience of training a robot policy and implementing it on real hardware. This practical exposure is the single most valuable thing an aspiring roboticist can do. When it comes to coursework, his recommendation might surprise some: focus on math. While many might expect him to suggest programming or robotics-specific courses, he argues that a strong foundation in linear algebra, optimization, and statistics is absolutely fundamental for a career in robot learning. As the field progresses and machine learning techniques abstract away more of the low-level details of the robot, a deep understanding of the underlying mathematical principles will become even more critical. He believes that this mathematical fluency is what will enable the next generation of researchers to develop truly novel learning algorithms. Of course, classes like robot dynamics are also very useful, but the core message is that a mastery of fundamental mathematics is the key to unlocking future breakthroughs in robotics and AI.

Future breakthroughs in household robotics. Looking ahead to the next five years, Jeremiah predicts that the most significant breakthrough in robotics will be the emergence of capable, general-purpose household robots. He envisions a near future where you can give a language command to a robot, such as “fold the laundry” or “load the dishwasher,” and it will be able to perform the task with a high success rate—perhaps 90% to 95%. While not perfect, this level of reliability would be sufficient to make these robots genuinely useful for a wide range of non-critical household chores, fundamentally changing how we manage our homes. However, he is quick to point out a major barrier to widespread adoption: cost. The current generation of advanced humanoid and mobile manipulator robots can cost tens of thousands of dollars, placing them far out of reach for the average consumer. He mentions the Stretch robot from Hello Robot as a notable exception, as its design philosophy was to become the “Roomba of mobile manipulation” by focusing on a simpler, more affordable platform. The existence of such projects suggests that there may be pathways to lower-cost solutions. Nonetheless, Jeremiah’s prediction is nuanced: while he is confident that the technology for useful household robots will be ready within five years, he remains uncertain about whether the price point will be low enough for them to become a common sight in our homes. The breakthrough will be in capability first, with affordability likely to follow later.

Looking forward to the final chapter. As Jeremiah enters the final year of his PhD, he is filled with a sense of excitement and anticipation. Naturally, he is looking forward to graduating and the opportunities that will follow, including traveling to conferences to share his work. However, his enthusiasm runs deeper than simply crossing the finish line. He expresses a genuine joy in the research process itself — the cycle of running experiments, collecting data, training policies, and then seeing his innovative ideas come to life in the behavior of a physical robot. This tangible feedback loop is a source of immense satisfaction and fun for him. Beyond his individual research, Jeremiah is most looking forward to embracing his final year within the collegiate environment at Georgia Tech. He plans to fully leverage the resources the university offers and, most importantly, to continue to collaborate with other people on cool projects. This desire for collaboration is a recurring theme in his journey. He sees his final year not as a solitary push to the end but as a last opportunity to be immersed in a vibrant community of peers, working together to solve interesting problems. For Jeremiah, the people and the shared intellectual journey are just as important as the research itself, and that is what he is most excited about as he completes this pivotal chapter of his life.

On the go? There’s an audio-only version too. Click here.

Jeremiah Coholich is a Robotics PhD student in the Robotics Perception and Learning (RIPL) Lab, advised by Zsolt Kira. Jeremiah’s diverse journey has included internships at NASA JPL, SpaceX, Harmonic Bionics, and the Honda Research Institute, as well as key roles in undergraduate research at the Human Centered Robotics Lab and co-founding Longhorn Racing Electric.