
cs & math @georgiatech ‘30 365 days of robot learning 50/365
Atlanta, GA
Joined January 2022
- Tweets331
- Following650
- Followers833
- Likes1.3K
Pinned Tweet
Day 32/365 of robot learning.
I got an so-101 arm to play ping pong!
The video was the first successful run, but it’s been averaging around 13 bounces with a high score of 23
I’ll make a write up on this soon, but in the mean time I’ll keep making the policy more robust!
Day 31/365 of robot learning.
Today marks one month since I started this challenge!
I'm still working on getting my so-101 robot arm to bounce a ping-pong ball using zero real-world demos, purely through RL trained in simulation.
It's not fully working yet, but I'm slowly making progress and solving one problem at a time. The arm can now launch the ball about half the time, bounce it up to 3 times in a row, and even managed to balance it on the paddle for 23 seconds once.
Most of the problems haven't even been with the RL algorithm itself, but rather getting the simulation to match reality. For example, the real paddle and ball aren't as bouncy as in simulation, and the servos are much slower under load than I modeled. Even getting the paddle perfectly flat was annoying since being off by just a few degrees would make the ball roll off.
I've also spent a lot of time debugging camera calibration, joint offsets, and servo dynamics. I even ended up using the ball itself as a level to calibrate the paddle.
Safety was another interesting problem, especially since I have to drop the ball onto the paddle by hand. I made the arm wait completely still until the camera detects the ball, rather than immediately executing the policy. I also trained with simulated hand disturbances, moments when the ball is hidden from the camera, and false detections to make the policy more robust. On the real arm, I've added bounded joint movements, smooth transitions, and automatic recovery when the ball is lost. I'm still working on explicitly detecting when my hand has left the workspace before allowing the arm to move.
I'm now retraining with a more realistic servo model to hopefully get it working soon.
Also, on a side note, I genuinely don't know how people online doing similar challenges manage to balance a normal work or college life with research and consistently working on side projects.
It's becoming pretty hard trying to balance research, open-source work, hackathons, classes, the gym, trying to go on a cut, and all the other responsibilities. But looking back, I've grown incomparably more over this past month. So I'm going to keep going for the next 11 months until I finish this challenge.
I definitely plan to keep going beyond the 365 days, but I also really need to figure out how to make all of this sustainable. Also thanks to everyone who provided me with resources and guided me along the way!
Day 49/365 of robot learning.
I spent most of today on WAM interpretability experiments and my computer organization homework. I’ve been varying different properties to see how closely the arm follows the future the model generates. What I’m really curious about is how stubborn it is about following that future, even when it’s a completely wrong scene or way out of distribution of training demos. If the arm still follows it closely across all those cases, I wonder if the main bottleneck to getting the policy to generalize to everything is getting the video model to generate the right future. If we could give it a future that shows the motion we want, how much could we steer the policy? I’m still running experiments, so I’m keeping the details light for now because I don’t want to scoop my own research idea.
On the side, I’ve also been working with a friend on training a VLA and trying to get it running as fast as possible on an FPGA. I’ve been getting more interested in robotics infrastructure because I think there’s a lot of room to optimize inference at different layers of the stack, even with the model architecture fixed. Inference optimization was something I was into before I got into robotics, so it’s been fun coming back to it.
I think I might be taking on a few too many projects lol. Over Thanksgiving break I’m planning to finally write up the ping pong RL demo. I also want to get moving on my humanoid build. So far I’ve put together the parts list and bought a few things. I still need to order the rest and start putting it together.
Day 48/365 of robot learning.
I started reading RAX today, and I really liked the Richard Sutton quote they started off with. “We want AI agents that can discover like we can, not which contain what we have discovered.”
I liked how they described the current state of robot learning. When we want to teach a robot a new task, we often collect a whole new set of teleoperation demos for that task. The policy learns from what we show it, but once something falls outside that distribution, it usually has a pretty limited way of figuring out what to try next. I think whatever approach we pursue should give robots room to discover things on their own.
RAX does this through exploration. In the air fryer demo, the agent used measurements from its own run to figure out how the drawer moved and wrote code to open it again in a new scene. The agent spent 72 minutes exploring, and the program it wrote could do the task in 18 seconds.
I’m still open on how far imitation learning can go as well, especially as these policies scale. I keep seeing these demos from Generalist, and I think it’s pretty cool how much GEN-1.5 can figure out through in-context learning. In one demo, they showed it two separate examples, one of unzipping a pouch and another of taking money out, and it figured out how to move from one to the other on its own.
Have a lot of exams and research these next few days, but will hopefully get through RAX as well over the weekend.
Praneeth Samineni retweeted
Our colleague Eva Yi Xie, a Princeton neuroscience PhD student, was detained by ICE at Reagan National Airport on her way home from SfN, where she’d just presented her poster. She’s being held in Chantilly, VA. Please help us bring her home. 📷
Day 48/365 of robot learning.
I started reading RAX today, and I really liked the Richard Sutton quote they started off with. “We want AI agents that can discover like we can, not which contain what we have discovered.”
I liked how they described the current state of robot learning. When we want to teach a robot a new task, we often collect a whole new set of teleoperation demos for that task. The policy learns from what we show it, but once something falls outside that distribution, it usually has a pretty limited way of figuring out what to try next. I think whatever approach we pursue should give robots room to discover things on their own.
RAX does this through exploration. In the air fryer demo, the agent used measurements from its own run to figure out how the drawer moved and wrote code to open it again in a new scene. The agent spent 72 minutes exploring, and the program it wrote could do the task in 18 seconds.
I’m still open on how far imitation learning can go as well, especially as these policies scale. I keep seeing these demos from Generalist, and I think it’s pretty cool how much GEN-1.5 can figure out through in-context learning. In one demo, they showed it two separate examples, one of unzipping a pouch and another of taking money out, and it figured out how to move from one to the other on its own.
Have a lot of exams and research these next few days, but will hopefully get through RAX as well over the weekend.
Day 47/365 of robot learning.
Today I finished reading ASPIRE and got a better sense of why it’s so expensive to run. When their iterative debugging stalls, evolutionary search lets the agent test several candidate repairs in simulation and use the best ones to guide the next round. This lets it explore different approaches, but testing those candidates adds up. In the 60-task LIBERO-Pro run with Opus 4.6, evolutionary search alone added about 17k LLM calls and 9.5k simulation rollouts across 31 tasks.
I also kind of related with the problem they encountered with resetting Rubik’s Cube demos. In simulation, you can reset the scene and sample a new starting state in code. On the real robot, a person still had to give the initial task guidance, scramble the cube, and reset the scene between trials. I did some digging and found that ENPIRE automates resets for specific tasks. For example, after a GPU insertion trial, the robot unplugs the GPU and verifies that the reset worked. I think making a framework that makes task self resetting more randomized to give the policy a better testing environment would help us benchmark robotics better and faster.
Another limitation I kept thinking about was what the agent can observe through the robot API. ASPIRE gives it tools for perception, motion planning, and control, and the execution engine records traces the agent can use to debug failures and turn fixes into reusable skills. When the API can’t give the agent a measurement it needs, it has to infer what happened from the observations it already has. If the gripper hides a cube face, for example, the agent might assume the turn completed and plan the next move from the wrong cube state. Giving it another view could let it check before continuing (this example is a bit basic but you get the point). For a jammed connector, force feedback and compliant control could help it detect the jam, back off, and adjust the angle. I think expanding the tools available through the API would progress agentic robotics faster.
Tomorrow I’m moving on to GR00T 2, then RAX.
Praneeth Samineni retweeted
Shitij has fantastic insight, is extremely motivated, and was an absolute pleasure to work with on this project.
He's applying for grad school this cycle - somebody get this guy a PhD!
Check out our work on representing “motion” as 3D scene flow with a video generation backbone. XYZ->RGB trick for using pretrained WAN VAE might be my favorite idea 😆
Learned so so much on this project, great collaborators and mentors: @jerthesquare_ @NamraByTheSea @agarwal_ayush05 @animesh_garg
Day 47/365 of robot learning.
Today I finished reading ASPIRE and got a better sense of why it’s so expensive to run. When their iterative debugging stalls, evolutionary search lets the agent test several candidate repairs in simulation and use the best ones to guide the next round. This lets it explore different approaches, but testing those candidates adds up. In the 60-task LIBERO-Pro run with Opus 4.6, evolutionary search alone added about 17k LLM calls and 9.5k simulation rollouts across 31 tasks.
I also kind of related with the problem they encountered with resetting Rubik’s Cube demos. In simulation, you can reset the scene and sample a new starting state in code. On the real robot, a person still had to give the initial task guidance, scramble the cube, and reset the scene between trials. I did some digging and found that ENPIRE automates resets for specific tasks. For example, after a GPU insertion trial, the robot unplugs the GPU and verifies that the reset worked. I think making a framework that makes task self resetting more randomized to give the policy a better testing environment would help us benchmark robotics better and faster.
Another limitation I kept thinking about was what the agent can observe through the robot API. ASPIRE gives it tools for perception, motion planning, and control, and the execution engine records traces the agent can use to debug failures and turn fixes into reusable skills. When the API can’t give the agent a measurement it needs, it has to infer what happened from the observations it already has. If the gripper hides a cube face, for example, the agent might assume the turn completed and plan the next move from the wrong cube state. Giving it another view could let it check before continuing (this example is a bit basic but you get the point). For a jammed connector, force feedback and compliant control could help it detect the jam, back off, and adjust the angle. I think expanding the tools available through the API would progress agentic robotics faster.
Tomorrow I’m moving on to GR00T 2, then RAX.
Day 46/365 of robot learning.
Most of today went into exploring papers as I slowly get more into agentic robotics. I added GR00T 2 and RAX to my reading list and kept reading ASPIRE after finally finishing AW-0 yesterday.
I also got to benchmark AW-0 against ASPIRE. In the cleaner scenes I tested, AW-0 matched or slightly beat ASPIRE. Most of AW-0’s failures showed up in cluttered scenes. Since AW-0 was trained on tidy scenes, I think it's partly because it's just OOD.
As I read ASPIRE, I spent some time getting familiar with the framework. I looked at how the coordinator assigns each task to a coding agent so multiple agents can work on different tasks in parallel. Each agent writes a robot program, runs it in simulation, and uses feedback from the rollout to revise it. I also spent some time exploring the robot execution engine and how agents use depth images, grasp candidates, and motion-planning results to figure out failure cases and address them.
Planning to finish ASPIRE next, then dig into GR00T 2 and RAX.
Praneeth Samineni retweeted
We used to hike with our robot, controller in hand.
Now it hikes autonomously, 10 km down from the Matterhorn. 🏔️🐄
We published a new paper in Science Robotics:
📄 science.org/doi/10.1126/scir…
🌐 roam-hiking.github.io
Praneeth Samineni retweeted
Check out our work on representing “motion” as 3D scene flow with a video generation backbone. XYZ->RGB trick for using pretrained WAN VAE might be my favorite idea 😆
Learned so so much on this project, great collaborators and mentors: @jerthesquare_ @NamraByTheSea @agarwal_ayush05 @animesh_garg
Praneeth Samineni retweeted
Most world action models learn to act by predicting videos.
We find that predicting how the scene moves in 3D performs better.
Introducing FloMo: a world action model that uses 3D motion as the bridge between video pre-training and robot control.
Day 46/365 of robot learning.
Most of today went into exploring papers as I slowly get more into agentic robotics. I added GR00T 2 and RAX to my reading list and kept reading ASPIRE after finally finishing AW-0 yesterday.
I also got to benchmark AW-0 against ASPIRE. In the cleaner scenes I tested, AW-0 matched or slightly beat ASPIRE. Most of AW-0’s failures showed up in cluttered scenes. Since AW-0 was trained on tidy scenes, I think it's partly because it's just OOD.
As I read ASPIRE, I spent some time getting familiar with the framework. I looked at how the coordinator assigns each task to a coding agent so multiple agents can work on different tasks in parallel. Each agent writes a robot program, runs it in simulation, and uses feedback from the rollout to revise it. I also spent some time exploring the robot execution engine and how agents use depth images, grasp candidates, and motion-planning results to figure out failure cases and address them.
Planning to finish ASPIRE next, then dig into GR00T 2 and RAX.
Day 45/365 of robot learning.
Today I finished reading AW-0. One of the things I found pretty cool was the hybrid execution setup. AW-0 can either predict actions directly through its learned WAM policy or execute generated code through a robotics framework with perception and motion-planning tools. At each replanning step, it checks whether the generated code passes validity and progress checks, otherwise it falls back to the WAM. On RoboTwin 2.0, WAM-only achieved 80.8% success in clean environments compared to 63.1% for hybrid, but under randomized conditions, hybrid performed better at 29.2% versus 20.6%. I wonder if this was just because agentic frameworks are better with OOD tasks. I think the two configurations also use different training demonstrations, so it's not a completely controlled comparison.
Tomorrow I want to look more into how AW-0's agentic pathway compares with ASPIRE, since both use code-based robot skills but approach the problem pretty differently. Will post any results I find tmw. There's also an interesting interpretability question here around whether the generated code actually provides meaningful planning information to the action policy, or whether the model mainly benefits from learning to predict code during training.
Praneeth Samineni retweeted
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me.
I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA.
I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation.
Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO.
Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours.
Jensen
Day 45/365 of robot learning.
Today I finished reading AW-0. One of the things I found pretty cool was the hybrid execution setup. AW-0 can either predict actions directly through its learned WAM policy or execute generated code through a robotics framework with perception and motion-planning tools. At each replanning step, it checks whether the generated code passes validity and progress checks, otherwise it falls back to the WAM. On RoboTwin 2.0, WAM-only achieved 80.8% success in clean environments compared to 63.1% for hybrid, but under randomized conditions, hybrid performed better at 29.2% versus 20.6%. I wonder if this was just because agentic frameworks are better with OOD tasks. I think the two configurations also use different training demonstrations, so it's not a completely controlled comparison.
Tomorrow I want to look more into how AW-0's agentic pathway compares with ASPIRE, since both use code-based robot skills but approach the problem pretty differently. Will post any results I find tmw. There's also an interesting interpretability question here around whether the generated code actually provides meaningful planning information to the action policy, or whether the model mainly benefits from learning to predict code during training.
Day 44/365 of robot learning.
Today I looked more into interpretability for world action models, and it almost feels like a goldmine with how much there is still left to explore. A lot of what I’ve been reading focuses on whether generating future video helps task success and how much inference time it costs. But there are so many properties of these models that those measurements don’t tell us much about.
I mainly spent some time exploring papers like FloMo, Tapestry, and FasterWAM, which take different approaches to using pretrained video models for robot control. Tapestry generates actions without generating video at inference, while FasterWAM explores efficient future conditioning. FloMo predicts 3D scene flow, representing how points in the scene move in 3D, instead of full future frames. With all these different approaches, it makes me want to understand more precisely what information each representation ends up giving to the policy.
For example, what information from the video model makes it into the action policy, and which parts does the policy use? How much do object motion, geometry, or visual details influence its actions? Even if information is present in a hidden representation, that doesn’t mean the policy relies on it. And a higher task success rate alone doesn’t explain what changed internally or why one approach generalizes better to unfamiliar settings. I’d like to see more benchmarks focused on understanding these internal representations and their influence on behavior.
There are more specific properties I’ve been thinking about, but I’m keeping those ideas private for now. Will share more once I’ve tested them.
Day 44/365 of robot learning.
Today I looked more into interpretability for world action models, and it almost feels like a goldmine with how much there is still left to explore. A lot of what I’ve been reading focuses on whether generating future video helps task success and how much inference time it costs. But there are so many properties of these models that those measurements don’t tell us much about.
I mainly spent some time exploring papers like FloMo, Tapestry, and FasterWAM, which take different approaches to using pretrained video models for robot control. Tapestry generates actions without generating video at inference, while FasterWAM explores efficient future conditioning. FloMo predicts 3D scene flow, representing how points in the scene move in 3D, instead of full future frames. With all these different approaches, it makes me want to understand more precisely what information each representation ends up giving to the policy.
For example, what information from the video model makes it into the action policy, and which parts does the policy use? How much do object motion, geometry, or visual details influence its actions? Even if information is present in a hidden representation, that doesn’t mean the policy relies on it. And a higher task success rate alone doesn’t explain what changed internally or why one approach generalizes better to unfamiliar settings. I’d like to see more benchmarks focused on understanding these internal representations and their influence on behavior.
There are more specific properties I’ve been thinking about, but I’m keeping those ideas private for now. Will share more once I’ve tested them.
Day 43/365 of robot learning.
Today I started watching MIT’s Flow Matching and Diffusion Models, but most of my time went into C programming for my computer organization class. I got pretty sick today, so it was a slower day. This course has been on my watch list for a while, and I’m glad I finally started working through it. With how often flow matching and diffusion come up in the robotics papers I’m reading, I want to get more comfortable deriving the math and understanding it.
As winter break approaches, my goal is to solidify my fundamentals while continuing research in parallel. Reading all these papers exposes so many gaps in my understanding, so I want to keep working to fill them in. I’m also working on ordering parts for my humanoid, so hopefully there will be more hardware updates soon.
On the side, I’ve been exploring tactile sensing with my friend @hemil_thakkar07. He’s handling the 3D printing and building, and we’re using a low-cost design with magnets inside a 3D-printed structure. We’ve been abusing the Georgia Tech makerspace for this. More experimentation on that soon. I feel like this is an area I have to explore because touch could help address so many deployment bottlenecks like plugging cables into server racks or adjusting a part that gets stuck during assembly.
Day 43/365 of robot learning.
Today I started watching MIT’s Flow Matching and Diffusion Models, but most of my time went into C programming for my computer organization class. I got pretty sick today, so it was a slower day. This course has been on my watch list for a while, and I’m glad I finally started working through it. With how often flow matching and diffusion come up in the robotics papers I’m reading, I want to get more comfortable deriving the math and understanding it.
As winter break approaches, my goal is to solidify my fundamentals while continuing research in parallel. Reading all these papers exposes so many gaps in my understanding, so I want to keep working to fill them in. I’m also working on ordering parts for my humanoid, so hopefully there will be more hardware updates soon.
On the side, I’ve been exploring tactile sensing with my friend @hemil_thakkar07. He’s handling the 3D printing and building, and we’re using a low-cost design with magnets inside a 3D-printed structure. We’ve been abusing the Georgia Tech makerspace for this. More experimentation on that soon. I feel like this is an area I have to explore because touch could help address so many deployment bottlenecks like plugging cables into server racks or adjusting a part that gets stuck during assembly.
Day 42/365 of robot learning.
Today I continued working through AW-0, focusing on how its three transformer experts generate code, future video, and robot actions together. I worked through the math behind how the code expert fills in hidden tokens using masked diffusion, while the video and action experts use flow matching to turn random noise into continuous predictions. Also how to represent clean data mixed with noise and how to derive the velocity the model learns to predict.
I continued my review of the standard transformer architecture from yesterday as well, mainly using active recall to rederive the functions behind how attention passes information between tokens and how the feed-forward networks update each token’s internal representation.
On the side, I’ve also been learning how probes work from the ground up for my WAM interpretability research, cuz before this I was blindly hoping Claude Code was doing it properly. Looked over how to keep the policy frozen, record its hidden states, and train a small predictor to test what information they contain. I think I'm going to look over interventions next.
I’m still rebuilding these foundations, but it’s helping me get more precise about what I understand and what I need to revisit. Idk yet if deriving all this math is contributing to my intuition, or if just the high level understanding is enough when we have a bunch of agents under us. I feel like my undergrad is the one time I can prioritize the learning so I'll do it either way.
Praneeth Samineni retweeted
100k+ signatures in less than 24 hours! 💚
Over 5k current NVIDIANs, nearly 1k former NVIDIANs, and 54k+ shareholders. Thank you all!
We hope the NVIDIA Board hears these voices and considers our requests.
And to Jensen: Thank you. We stand with you. 💚
#standwithjensen #bringjensenback
standwithjensen.com/
Today, we enable AutoResearch in the physical world for the first time! Introducing ENPIRE: we give 8 Codex agents a fleet of robots, an allocation of GPUs, and generous token budget. We set them free with a simple goal: solve the task as quickly as possible, keep the robots busy but stay safe, don't waste precious compute. Make no mistake.
Then humans step aside and our watch begins. The robot fleet starts to come alive: they learn to look for visual clues, reset the scene, practice novel skills, tinker with control stack, read papers online, debate, reflect, get stuck, and try again directly on the hardware. All we did is to give Codex an API to the world of atoms, and the rest is emergence.
ENPIRE is able to solve high-precision tasks like tying zip-ties, organizing fine pins, and installing GPUs all by itself. We also discovered a new type of "physical scaling": 8 robots exploring in parallel improves significantly faster than fewer ones.
A part of our NVIDIA GEAR lab now self-improves tirelessly over night. We just read the reports in the morning.
/goal: we all take a holiday and Jensen wouldn't even notice ;)
We will be open-sourcing everything, so you can host your self-running robot lab at home too! Deep dive in the thread:
Day 42/365 of robot learning.
Today I continued working through AW-0, focusing on how its three transformer experts generate code, future video, and robot actions together. I worked through the math behind how the code expert fills in hidden tokens using masked diffusion, while the video and action experts use flow matching to turn random noise into continuous predictions. Also how to represent clean data mixed with noise and how to derive the velocity the model learns to predict.
I continued my review of the standard transformer architecture from yesterday as well, mainly using active recall to rederive the functions behind how attention passes information between tokens and how the feed-forward networks update each token’s internal representation.
On the side, I’ve also been learning how probes work from the ground up for my WAM interpretability research, cuz before this I was blindly hoping Claude Code was doing it properly. Looked over how to keep the policy frozen, record its hidden states, and train a small predictor to test what information they contain. I think I'm going to look over interventions next.
I’m still rebuilding these foundations, but it’s helping me get more precise about what I understand and what I need to revisit. Idk yet if deriving all this math is contributing to my intuition, or if just the high level understanding is enough when we have a bunch of agents under us. I feel like my undergrad is the one time I can prioritize the learning so I'll do it either way.
Day 41/365 of robot learning.
As I’ve been working through AW-0 and DreamZero, especially the mixture-of-transformers architecture in AW-0, it’s been hard to develop an intuition because my understanding of transformers and diffusion models was pretty shallow. Most of it came from Andrej Karpathy and 3Blue1Brown videos I watched more than a year ago, so today I went back and reviewed the basics and started rederiving the math for the different steps using. I just started today, so I’ll be doing the rest over this weekend.
As I explore these different architectures, it also makes me question how much different parts of them affect the policy, which I’ve been exploring in my WAM interpretability research. I think I’m going to spend a few days soon exploring probes, K/V swaps, activation patching, ablations, and activation steering to be able to experiment more with these architectures. I'm also finally done with exams, so I'll be ramping up on side projects over the next few days!
Praneeth Samineni retweeted
Okay, here's one thing that's been bothering me after attending #CoRL2026. 🧵
To me, whether to work in a given area is such a personal choice, made by individual scientists depending on their research taste and idiosyncrasies.
The best part about science is that we don't have to agree!! I can think that Area or Problem X is not worth working on, and anyone can (and probably will) disagree with me and end up proving me wrong with evidence.
To me, science involves trying to achieve consensus on (ideally falsifiable) hypotheses *after* conducting research. Not before. Debates are very useful and necessary for this.
But perhaps we should think more carefully about claims like "Area X is not worth working on". I heard many of these being freely dispensed at this past week.
I feel that these subjective claims are nearly impossible to verify. More importantly, I worry that such sweeping claims may cause many junior researchers working in that very area to feel considerable distress.
Note that this is very different from lobbying for everyone to work on our pet problems by highlighting their motivations and potential impact. Such lobbying has long been a part of science, and will likely always be.
Maybe we can keep science weird and highly entropic! The very diversity of our idiosyncratic research agendas is what helps increase the odds that at least some of us will eventually identify a path to our ultimate destination — knowledge!




















