4 ms·

ROS is shit. In every possible way. It's a set of semi-related components, most of which can be done far better by doing it the normal way or writing it yoursel

by dima55 2y ago

ROS is shit. In every possible way. It's a set of semi-related components, most of which can be done far better by doing it the normal way or writing it yourself. Nobody wants a "DDS" to simply send bits across and nobody wants their build system and nobody wants their half-assed protobuf reimplementation. And so on.

causal 2y ago
I also have strong feelings of frustration towards ROS but I have yet to see a viable alternating. Half the value of ROS is the community building robot specific packages on the ROS foundation. Any frustration with the system is certainly smaller than rolling a new one.
- chfritz 2y ago
  Well said!
- a_t48 2y ago
  I'm working on one right now, no public code yet - http://basisrobotics.tech http://basisrobotics.tech (well, we - myself and cofounder) It's serialization independent (integrating a new serializer is super easy, you can even use ROS messages if you have a lot of legacy code to migrate), works with regular old CMake, and provides a framework to run deterministic replay/simulation against. Hoping to put up a Show HN myself in a few weeks - need a few more pieces in place first - I have the foundation of the replay system written, but it's not all there, yet. (Hi Christian!, Kyle here)
a_t48 2y ago
ROS is really great for prototyping. Agree that it has some major issues for production.
lowq 2y ago
I hear you man. The fact that a `byte` in ROS's bootleg protobuf is actually signed, and `string` is ASCII-only means that if you want to send a file or binary blob you have to use a `uint8[]`, which of course C++ will turn into a `std::vector<uint8_t>`. Good luck turning that into a proper `std::string` without an extra copy. And don't get me started on the comically scattered documentation for any basic use case that only ChatGPT can seem to make sense of. Better yet, the complete insanity that is node parameters - ever heard of reading from a file? Can't believe anyone takes this seriously. ZeroMQ, protobuf, and a little thought would go a long way at our org. /rant

by dima55 2y ago

causal 2y ago

I also have strong feelings of frustration towards ROS but I have yet to see a viable alternating. Half the value of ROS is the community building robot specific packages on the ROS foundation. Any frustration with the system is certainly smaller than rolling a new one.

chfritz 2y ago
Well said!
a_t48 2y ago
I'm working on one right now, no public code yet - http://basisrobotics.tech http://basisrobotics.tech (well, we - myself and cofounder) It's serialization independent (integrating a new serializer is super easy, you can even use ROS messages if you have a lot of legacy code to migrate), works with regular old CMake, and provides a framework to run deterministic replay/simulation against. Hoping to put up a Show HN myself in a few weeks - need a few more pieces in place first - I have the foundation of the replay system written, but it's not all there, yet. (Hi Christian!, Kyle here)

a_t48 2y ago

ROS is really great for prototyping. Agree that it has some major issues for production.

lowq 2y ago

I hear you man. The fact that a `byte` in ROS's bootleg protobuf is actually signed, and `string` is ASCII-only means that if you want to send a file or binary blob you have to use a `uint8[]`, which of course C++ will turn into a `std::vector<uint8_t>`. Good luck turning that into a proper `std::string` without an extra copy. And don't get me started on the comically scattered documentation for any basic use case that only ChatGPT can seem to make sense of. Better yet, the complete insanity that is node parameters - ever heard of reading from a file? Can't believe anyone takes this seriously. ZeroMQ, protobuf, and a little thought would go a long way at our org. /rant